diff --git a/.agents/skills/augustus/SKILL.md b/.agents/skills/augustus/SKILL.md index a9e6a94..c858696 100644 --- a/.agents/skills/augustus/SKILL.md +++ b/.agents/skills/augustus/SKILL.md @@ -54,7 +54,7 @@ classical method you already trust, substitute it, classify the win "paraphrase brittleness", "allowlist then judge", "TOCTOU-of-Noul", "Jev inside the database / sqlite-jev", "Jev picks bitrate / join order / the model", "wait for Archer", "lint the request / missing - other", "training confronts Choice other / none-of-the-above", "soft AGENTS.md rules vs the linter", "screenshot Choice / omni System One", "extractive quotes / pointer not generator", "compaction summarize vs pointer", "encoder vs Jev compaction backend", "shadow-mode compaction rollout", "CI flaky-vs-real merge gate", "fail-open VOI wake/resume", "claim vs session evidence", "S1 indexer escalate-S2", "Harbor on/off routing", "fail-open vs fail-closed wake vs CI gate", "encoder vs Jev computer-use backend", "hybrid local decide + remote fill", "DONE vs verified success", "stdout prune vs session compaction", "OpenCode jev-pruner vs Claude jev-pruner", "zen-chat vs jev-zen Noul", "hard envelope then Noul prune", "Cua-S1 vs TypeSafe Jev", "plan vs execute dry-run", "specialist computer-use vs general agent", "local drop-in vs stub scorer", "route vs memory", "when does it hold / extractable from state", "decision model vs constrained LLM", "dual-process S1/S2", "combinatorial grid vs extractive", "uncalibrated local likelihoods", "decision-native RAG", "classify-first / read selectively", "living applied-mappings atlas / class patterns", "silence as safer / draft-gate heartbeat", "robotics text-state vs pixels", "verbatim ledger vs summary", "judgment as language primitive", "Stagehand extract pick-and-copy", "harness observe-score-act vs demo loop", "public judgment wall / six parallel questions", "meaning-search without embeddings", "attention ≠ correctness", "skills→oxlint / AST prove ∩ remainder", "session-sticky first-prompt routing", "measured RAG rerank vs generative rerank", "capability kernel / secrets never in the agent", "Jev is SENSOR not policy", "type-safe ≠ correct", "typed control plane around DSPy", "native vs verbalized confidence", "engine owns truth / Jev owns judgment", "human-confirmed kill gate", "train specialist vs few-shot hosted", "decide→policy→LLM leftover", "Noul 0.5 cannot-tell never rounded", "calibration ≠ sortable / ORDER BY", "pairwise inversion / Score ordinality / two-decimal ties", "wire-compat GLiFormer /v1/systemone", "class-backend economics", "loopback gateway hosted + local", "do not distill Jev as teacher", "active-learning triage", "evidence-packet explorer", "meaning-grep AND/OR/NOT", "closed-vote-only / no planner LLM", "Jev vs PCD Harbor", "PCD O(1) ≠ Noul", "host-owned handlers × System One", "OMP/pi fail-open gate", "permission vs probability / operator owns thresholds", "judgment ≠ permission / Jev never grants access", "eval integrity / instrument not score", "constrained optimizer + S1 features / never sole hot-path gate", "privilege ≠ verdict / effect contracts not tokens", "attention filter / VOI for human review / never blocks / never green unless sure", "measurement owns endorsement / evidence-gated question packs", "Jev supplies evidence / code owns authority", "ranking ≠ calibration / never hard-threshold raw p as frequency", "hot-click CU / indexed element table", "Jev judges relevance / code decides structure", "local rules first then remainder / never auto-train on own hides", "combinators / System One as control plane", "receipts not leaderboard / type-safe ≠ correct jaggedness", "VOI over skill library / skillranker abstention", "OOD calibration / AUC ≠ ECE", "Jev vs thinking-budget small models", "turnstile / replayable evidence≠authority", "MLX one-pass schema→JSON / Apple Silicon replica economics", "memory leases ended by new evidence", "never confidently wrong / TLA+ compose / escalate instead of hard-gate", "no seal no advance / coverage ledger / mint ≠ product brain", "skill-broker sibling / judgment ≠ permission", "sureness bands / max_prob is generous", "JevBench / calibration not in Main Score", "CI typed gate before expensive review", "Codex MCP host adapter", "judgment as attention redirect / jev-preflight", "compress-before-first-send / dizk jev-lens", "tools≠use / SessionStart over hoping", "observational memory / pi-om keep-kind", "open-Jev class / openvons / JevPick", "physical-world System One / HA-Jev / not for locks", "judgment outside the store / jevql", "landed-script trust / headless≠auto-approve", "digital-design combinators / extended five", "VOI cache admission / same-intent skip LLM", "BM25 vs Jev skill routing Harbor harness", "zeroshot vs BERT / contamination DiD", "typed escalate continue abort baton / inverted loop", "worth-your-attention VOI / ThinkyMiner Winnow", "Jev WHETHER Python HOW LLM WHAT", "conflict vs ignorance / named Choice escape", "Playwright executes Jev chooses", "OpenJev /v1/decide not drop-in", "SemIf wire-compat runoff; SemIf rename densify / MLX backend / 5.21× systems≠semantic / Softmax ≠ Noul (`notes.md` §117)", "decision-as-memory flywheel", "record/replay CI / jevassert", "failure-finding arena / jevarena ≠ jev-arena", "BBQ not a bias cert", "decider≠executor", "sentence-as-rule lint / jevlint", "sentence-as-rule lint / jev-lint is jevlint rename", "VOI hunk prune", "whole-repo intent VERIFIED/VIOLATION/UNKNOWN", "GLiNER2 spec ≠ replica", "open replica substrates / grande / laya-jolt / JEV-CPU", "ONNX local-jev not equivalent", "persist constraints across compaction / pi-heed", "calibration+cost first-class gates", "Harbor-shaped Jev vs SGR LLM-as-judge / jev-judge-bench ≠ jevarena ≠ jevbench", "hand no-text steps / jev-use / Vercel drops confidence", "Pi System-One control plane / pi-jev-control", "never free-generates / jev-gpt tree of Choices", "OpenRouter recipe atlas / samples not benches", "personal history feed / jevfeed / no social graph", "competing NAR claims / dual-channel ECE / openJev-verdict ≠ OpenJev", "empty compaction-proxy skip / IPECTER", "throughput ≠ latency / like-for-like ECE", "1-token logprob endpoint ≠ Noul / coverage ≠ correctness", "open replica engine / jevinf", "unofficial Elixir SDK ≠ OTP peer", "jevex n=16 files-to-read VOI", "commit pre-review attention≠verdict / middle band", "Hermes plugin is Agnes not TypeSafe", "pi-jev-compact ≠ pi-jev-compaction", "empty Codex-proxy skip / IPECTER runway", "decision-native inbox / mailordinal", "unofficial jev-cli not ready / ≠ jevql", "laya-multilingual / English checkpoint confident-wrong OOD", "schema-scorer peaked ranking ≠ calibration", "HF 401 / GitHub 404 Hub-only", "productized System One HTTP / classifier.dev", "escalate-under-threshold / smart tier / multi-label ignores", "silent FALLBACK / granite 0.546 vs advertised 0.800", "vs_jev tracked JSON / read eval/README", "choxos/jev-reviewer ≠ egma-ai / systematic-review pointer", "two-pass Choice+Noul evidence extraction", "not-found is an answer", "human check as productized judgment", "githubnext/localjev ≠ kunchenguid/local-jev", "wire-compat ≠ logit-equiv / prompted JSON ≠ structured read", "self-reported probs / entropy confidence", "GitHub Next local /v1/systemone", "LM Studio runner gap / structured-read primitives", "NandhaKishorM/laya packaging ≠ Hub-only / Router script-before-p", "post-T ECE ≠ raw ECE / Banking77 token-budget", "0.85 still soft / not TypeSafe drop-in", "external census ≠ scored bake-off", "GLiNER2+routers class-boundary", "incomplete openjev census vs watch", "Harbor honesty watch / silent fallback", "JevBench v1.2 geometric mean / cal ON rank / weight sensitivity", "option-order 72→21 / instruction models in the class table", "self-host latency ×2 assumption / est. costs", "Laya absent is a gap not a named exclusion", "Qwen3.8 27B ≠ Archer", "hourly already-folded watch / apply-the-five / skip thin noise", "hard-gate Noul as PR/quality gate is soundness theater", "S1 never stalls waiting / S2 one-use advisory", "Local controller ≠ githubnext/localjev", "purple telemetry = consumed not arrived", "seed = geometry not async replay", "20% starting gate still soft", "no pixels to either provider", "OCR+AX observe-score-act / typesafe-computer-use", "never send screenshot to frontier for the decision", "overlapping CU options = false low confidence", "split kind/item/site", "155× one-screenshot ≠ Harbor taskset", "decision ≠ answer-reader capture", "ASR observe-score-act / jev-voice-browser", "partial-speech VOI / free-text waits", "spoken confirm ≠ hard auth", "numbered overlay without another model", "wrap-as-execution / AgentGhost ALLOW ASK DENY", "rules first then Jev remainder / ASK throws / fail-closed", "reddpy/AgentGhost ≠ jwen5419807/agentghost ≠ vventirozos", "JP genre atlas / studio_yebisu / stars ephemeral ≠ eval", "Jev Clearly Explained / akshay_pachaar / LLM hammer", "schema-safe ≠ correct / 200× 400× TypeSafe ceiling", "questions-as-code / shadow first / not a TypeSafe how-to", "proposition ≠ embedding / contrast-set", "boolean composition of soft Nouls / AND OR NOT", "uehaj/jev-semgrep ≠ semgrep.dev", "meaning-grep dedicated fold / not a gate", "decision-validated UI / Jev never authors text", "decision-as-assert / jevtest ambiguous band", "typed decisions drive UI / jev2ui", "hybrid S1 closed verb menu / anima3", "pointer-not-generator search / JevFind", "jev-frontier-bench ≠ frontier-100", "product bakeoff ≠ architecture duel / GLiClass", "four engines same questions / majority floor", "authorship named escape / not evidence", "ha-switchboard HA remains execution", "n8n classify/route/score / Low Confidence", "fast-jev-compaction-pi ≠ pi-jev-compact ≠ pi-jev-compaction", "jevloop full-distribution optimizer / no LLM in the loop", "laya-vision SmolVLM / score untrained", "Cerebellum-2B /v1/decide ≠ TypeSafe / wire-compat vs agent-routing", "laya-grounded not drop-in / Platt not temperature", "GestaltLabs/Jeff-1 ≠ logan-markewich/jeff / acc vs ECE n=9730", "stanley-code empty findings ≠ approval / human promote", "findme ≠ JevFind / NL memory beam-search FS", "jevsubrouter price workers not conversation / counts ≠ dollars", "feelings .feels() default 0.5 is Noul-0.5-never-rounded / ≠ hunch ≠ Probably", "apa-agent-harness ≠ AntonioCoppe/jev-harness / unpublished npm", "grok-bot-jev skill cannot force a bot that ignores it / A/B proxies not tokens", "Essentiel-Jev never authority / human every action", "enzo-mcp independently falsifiable claims / ≠ jev-sift", "pigeonhole OTHER skip / decision-as-filing", "jev-reliability Nothing about accuracy", "clduab11/jev-test ≠ realZachi/jevtest / Nothing runs yet", "jev-rag-benchmark Jev wins is not an assumption", "dairui1/jev-lab ≠ BrendanH18/jev-lab", "jevmail gmail.readonly / mailjay archive/trash", "ZHUBoer/ego-jev reserved __none__", "runWorkflow completed ≠ success", "jsort scores are relative", "Noul not Choice for scale", "groundedness-judge-bench native vs schema-guided", "implicit_true included in yes", "jev_playground 0 promotions", "routing-backtest 0.0447%", "yuyang2230/jev-agent-skill jev-1.13-free", "jev-techstack-classifier stack_config.json", "s1_ruby collapse late", "undecided? abstain", "2389-research/judgement license null", "confidence ≠ winner p", "typesafeai-sdk-community not a new species", "tpellet/hunch exit 3", "never-execute list", "jev-file-search scores not calibrated accuracy", "jev-linkmap Jev never sees S2 prose", "muhammedilyasy/jev-mail metadata only", "tidy none-of-folders stay", "tab-bouncer pinned/audio/current never closed", "lkclean Show fail-open", "jev-yt-time-saver Show anyway", "ORIGIN pause-if-no-Jev", "validResponse sums-to-1", "jev-crawlers risk bands never raw boolean", "jevbrain AUTO_ACT is not a Noul", "judgekit YAML classify/score/route/verify", "typed-judge-kit verdict-in-code", "alsoleg89/decide packing VOI", "0.8 ≠ 80% accuracy", "Jev-Calibration Platt ECE 0.117→0.052", "jev-calibration-arena never acts", "ctmx/openrouter-jev-mcp Decision-as-Plugin", "FrancoisChastel/jev-code ≠ npm jev-code", "claudecode-jev-marketplace fail-open not hot path", "pedroknigge/mcp_jev packs not ask_jev", "cyrusasco/typesafe-mcp noul deadband 0.35–0.65", "codaaiteam/jev-skill jevtypesafeai.com ≠ TypeSafe", "hermes-switchyard ≠ hermes-jev-router ≠ hermes-plugin-jev", "nanoprune 2.8MB ECE 2.58%", "smartdio/jev-browser-agent ≠ ZHUBoer/ego-jev", "Dakai/omp-jev-web DONE ≠ proof", "hari007sh/jev ≠ dannote/jev", "0thernet/system-one-skills deterministic verify", "typed-gate band [0.40,0.60] is refusal", "pi-jev-gate fail-closed; choice is the verdict", "Foq ~25ms/2.2GB local", "rev prefill-only + HF jev-0.5b", "robfrase/jev planning memo", "typesafe_agent_gates 27/27 / 31/31", "EpicEric/safe-sh static remainder", "pastepilot Confirm before act", "Jev-Reranker live Jev not yet measured", "sessionwise opt-in relevance", "jev-search pointer sieve", "400ms Salesforce WebMCP", "typesafe-scheduler-diagnostics advisory", "droidjev screenshot-free", "Tewoto1 jevcu planner still writes", "ha-conversation-jev Jev→Grok", "dsh-jev can only gate", "jev-classification-benchmark specified not run", "jev-luna-pagerduty p≥0.50", "meldltd/meldecision laya-go ONNX", "laya-doom never pixels", "logixism/laya-api empty README", "akpsahan/laya ≠ Archer", "choxos/jevchess engine owns truth", "jev-drive sim not AV", "story-arc Jev never authors", "jev-hs-assistant HS6", "golergka/jev-plays-starcraft-2 UI-verified ≠ API Victory", "awesome-jev-use-cases catalog", "Nibir1/typesafe-go ≠ official", "fingerprint after redact", "recall vs decide", "publish fingerprints+answers", "CI replay as Harbor cousin", "Cache hit ≠ correctness", "hyperspaceai/jevcache ≠ kushals256/jevcache", "human labels only", "score never auto-accepts", "production capture flywheel", "sutro-sh/jev-align ≠ caiovicentino/jev-align", "guidance ≠ hook", "catalysts ≠ summaries", "compile-time System One", "unofficial ≠ TypeSafe", "format_version modernbert-jev/1", "Argos1111/jev_local ≠ us/jev-local ≠ kunchenguid/local-jev", "LFM default ≠ ModernBERT backend", "Nemotron ≠ TypeSafe Jev", "not a calibrated replacement", "djev-dev complements djev-spark", "images as Choice options", "Laya essay numbers *theirs*", "Router/OOD confidence", "hosted bootstrap ≠ silent TypeSafe", "difficulty + policy thresholds + JSONL trace", "jev-codex-pilot model + reasoning depth", "keep/shadow/hybrid/reject", "quarry evidence projection", "Frank-ZY-Dou/awesome-jev robotics/3D/control", "one-dollar-tahoe TypeSafe Jev defense eval", "jevguard calibrator/cache/escape", "jev-ci-selector CI shadow mode", "llama-jev llama.cpp replica", "petercr/jev-orchestrator ≠ FleeexCorp/jev-orchestrator", "seb4ez/jevguard ≠ AseemPrasad/JevGuard ≠ pablozr/JevGuard", "webNeat/llama-jev ≠ WiktorB2004/llama-index-jev", "OpenCode jev-pruner context sieve", "observe→score-candidates→prune", "jev-zen / jev-1.13-free", "zen-chat ≠ Noul", "fail-open original", "keepScore >0.1 floor", "host port of tamaratran/jev-pruner", "indiejoseph/opencode-jev-pruner ≠ nrdz-labs/fast-jev-opencode", "jev-webagent-bench empty stub", "Kiln-AI/jev_jsonschema noul_threshold 0.5", "NSStudent/JevSwiftSDK unofficial", "GLiNER2 native Apple path", "unofficial Swift/Core ML GLiNER 2.5-small", "entity spans + confidence", "not Choice/Score/Noul", "not TypeSafe", "label descriptions as schema", "on-device ANE economics", "honesty locks", "shershah1024/gliner-native-runtime ≠ Fastino", "≠ gliner25-compaction ≠ gliner2-ultrafast ≠ Eran-BA/Jev_from_GLiNER2 ≠ NSStudent/JevSwiftSDK ≠ jevmlx", "default threshold 0.1 still soft", "soft Noul ≠ hard safety", "Decision Graph Protocol frame→assess→commit", "app retains permissions/effects", "Jev-first assessor-neutral", "guarded commit / receipt/next frame", "assessment batching", "hard-gating DGP as safety theater", "numerous-com/dgp ≠ TypeSafe official", "jegrep calibrated path+range Nouls", "no embeddings/index/daemon", "~$0.01–0.03 typical", "agent --json", "can1357/jegrep ≠ Bentlybro/jevgrep ≠ uehaj/jev-semgrep", "Archer-arch fidelity", "kev family OOD 0.76–0.77 vs Jev 0.86", "block-causal isolation", "pointer/readout CE-trained", "/v1/systemone drop-in", "replica honesty", "cost-sensitive decision theory × System One probabilities → control flow", "thresholds derived from costs not hard-coded", "YES / NO / UNSURE from cost_false_yes / cost_false_no / cost_human", "auto-batching same-object questions", "Kungie/gut ≠ tpellet/hunch ≠ carldaws/hunch", "judgment vs generation", "deterministic execution after probabilistic judgment", "exactly one app-owned callback", "explicit uncertain branch", "Illusion47586/judge ≠ lexingtonhibiki/judgekit ≠ Ascurse/typed-judge-kit", "variable-N option scoring as the trainable object", "dynamic candidate bags not fixed label sets", "zwliJay/jev-forge ≠ NanoJev", "open replica economics / latency vs closed Jev", "NAR local drop-in", "wfzyx/von late-catch HIGH", "competing NAR claims / replica honesty", "typed judgments vs chat judges on guardrailing", "ishaannk/llm-vs-jev cross-note only", "deeper integrity fold is rh-guard", "nothing wins outright", "can be argued out of guarding"", "Jev IS the if-statement", "judgments/probabilities drive branches", "text model only writes prose", "interpreter owns variables/loops/budgets/replay", "otherwise maybe / confidence gate", "chaos samples after the gate", "southpolesteve/probably ≠ carldaws/hunch ≠ feelings ≠ Kungie/gut ≠ Illusion47586/judge ≠ tidymodels/probably", "133★ / forks 10 live", "build calibrated classifiers from human feedback", "retrieve by relevance not resemblance", "one calibrated yes/no per memory in one request", "pointer mode 17/18 19/20 *theirs*", "embedding resemblance misses the allergy", "samdotmak/jev-recall ≠ jev-search ≠ jev-sift ≠ carryforward ≠ chopratejas/invalidate", "memory leases ended by new evidence", "six Nouls then fixed rules in code", "0 of 157 false invalidations", "questions/plans/directives are not evidence", "unsure → review queue", "host keeps the store", "name↔body / comment truth / test-claims", "mizchi/jev-lint is mizchi/jevlint rename", "no shipped rule has severity error", "~1 in 5 findings wrong *theirs*", "mizchi/jev-lint ≠ huntedman/JevLint ≠ MichitoSugawara/jev-lint", "JSON Schema → typed JSON via Jev", "noul_threshold 0.5 decoder not a proof", "IncompatibleSchemaError lists every bad property", "on-device Laya CoreML ANE", "~5 ms P50 short decisions", "189/189 FP16 checkpoint parity", "10× not achieved", "mizorewww/laya-coreml ≠ gliner-native-runtime ≠ jevmlx ≠ NandhaKishorM/laya", "softmax over allowed tokens ≠ Noul", "question-first cache", "Micha0827/snapjudge ≠ githubnext/localjev ≠ jevmlx ≠ cendress/SnapJudge", "Jev-first Pi agent loop", "slow-LLM fallback", "explicit action menu / CandidateSource unimplemented", "62 tests wiring not quality", "direwolfiy/JevPi ≠ standardagents/jevpilot ≠ pi-jev-control", "resume-screening bias audit methodology", "name×resume factorial independent Nouls", "callback determined by resume quality", "mean-probability name gaps operationally negligible", "natemoo-re/bias-bench ≠ BBQ", "Plan/PRD panel → code-owned pass|review|block", "cheerleading out of scope", "austindixson/planalyzer ≠ single-goodness Noul", "cost-aware multi-model routing/escalation", "decide vs do", "successful-task cost", "cannacre8ive/switchboard-ai ≠ ha-switchboard ≠ hermes-switchyard", "frozen-protocol zero-shot bench", "TypeSafe Jev vs PrismNLI vs Laya", "contamination caveat", "elcronos/jev-vs-open-decision-models ≠ JevBench ≠ DMB", "context-window admission control", "VOI gate which tokens are worth the expensive model", "fail polarity per lens", "on small inputs lenses lose money", "cvsgireesh/jevusher ≠ jev-sift ≠ winnow", "typed decision control plane", "receipt ≠ authorization", "historical-v0 zero retained cases", "MokiMeow/jev-fabric ≠ jev-forge ≠ dgp", "live 15-dim typed rubric re-score per pause", "scoring economics exemplar", "OpenJev/Codiv ≠ TypeSafe hosted", "jose-troche/live-rubric ~$0.000004 desc / ~$0.000006 README", "adversarial pre-registered Jev eval", "28 predictions before data", "123,805 requests", "confidence does not track ignorance", "polite injection 65% / crude 0%", "willkelly/jev-evaluation ≠ jevals ≠ jev-baselines-eval", "provider-neutral Elixir/BEAM Noul/Choice/Score SDK", "class infrastructure", "nshkrdotcom/system_one_sdk ≠ typesafe_sdk ≠ dannote/jev", "question-linting of Jev questions themselves", "nine jaggedness rules, no API key, no labelled data", "static lint ≠ measured separation", "yodablocks/jevq ≠ tenbin ≠ JevLint ≠ commitjev", "open-weights Laya as class exemplar (binding)", "Nx/Bumblebee runtime", "host chooses backend", "ChristianAlexander/laya_ex ≠ system_one_sdk ≠ dannote/jev ≠ NandhaKishorM/laya", "on-chain/edge Laya deploy", "parity_verified stays false", "model output never grants Tx", "humandebri/IC-Laya ≠ laya_ex", "auditable weekend replica", "Jev outputs never used for training", "soft human-vote distributions", "unpaired 0.577 vs 0.727", "agilabs-ai/jev48 ≠ JevBench ≠ Mapika/decider", "adversarial dual-judge / framing attack surface", "comparative framing is the usable judgment", "prior injection crowds out evidence", "copyleftdev/ember ≠ ember.js", "Laya specialist fine-tune pipeline", "training still GPU-pending", "PIXELZX0/XERON ≠ convaiinnovations/laya", "Hub Laya replica drop", "daliborsb/laya ≠ convaiinnovations/laya ≠ NandhaKishorM/laya", "System One student distillation corpus", "gold is programmatic", "teacher is closed-API clone", "do not distill Jev as teacher of record", "MagaBitmex/jev-4b-distill-data ≠ missing student checkpoint", "non-LLM VIN System One", "planning depth not chat", "lewislululu/jevon ≠ douglance/jevon", "source-bound evidence checks", "local quote mismatch needs no API", "exit 0 ≠ claim truth", "WaynezProg/jev-kit ≠ jonathanavis96/jev-kit (Airlock) ≠ jev-use ≠ jev-mcp", "independent System One evidence catalog", "scores not one leaderboard", "no external record currently reproduced", "TokenTrim no-Jev matched hybrid 62.4%", "reachjalil/system-one-bench ≠ mallahyari/system-one-benchmark", "21 tasks · 134 items · 208 questions", "scenes from public GitHub contracts, not production logs", "SivletLabs/jev-eval ≠ willkelly/jev-evaluation ≠ 4esv/jev-eval ≠ xxkuboxx/jev-eval ≠ onlyoneaman/jev-eval ≠ dayhaysoos/jevals", "option isolation (sibling-blind)", "permutation-equivariant", "Hub OWNER not published", "nafisazizir/hev ≠ jaredpalmer/kev", "frozen local LLM logits, no trained decision head", "residual-head 9,222-param decreased 73/96→67/96", "confidence = 1−normalized entropy, not P(correct)", "yuki-oshio/mini-jev ≠ r-ms/mini-jev", "Jev classifier as autoregressive next-token predictor", "ChatJev-style soundness theater", "erik-dunteman/ChatJev ≠ dannote/jev ≠ jev-gpt", "calibrated decision head × AlphaProof value head", "implementation-layer isomorphism, semantic difference", "timeout = censoring", "do not launder Noul as proof", "parallel rank-prediction vs serial selection", "independent questions can conflict", "zzzzzec/jevsort ≠ keltokhy/jsort", "curated open System One ecosystem catalog", "rupeshpoojary9/awesome-open-system-one ≠ AnotiaWang/awesome-jev", "arXiv paper radar with Jev relevance scoring", "ranking ≠ calibration / 0.5 still soft", "fail-open failed evals not marked seen", "train calibrated ~27M from scratch", "typed Q→prob dist / one forward pass / no LLM decode", "hyusi2003/MiniSystemOne ≠ Colvin0315/MiniSystemOne", "description-only stub / size 5", "ESCI hard probe fails four of six", "jev_bool ECE 0.242 inversion 0.255", "do not re-fold §60 six-gates as new", "jobbyjev one-request-per-company from batch-size result", "find/design/evaluate TypeSafe Jev decision loops", "karanb192/jev-architect ≠ samtay32/jev-system-architect", "Jairik/jev-distiller size 1", "distill-Jev UI stub / do not distill Jev as teacher of record", "post-launch scored use-case map / Jev self-scores then human curation", "licensedsaucer9-web/jev-opportunities", "Jev-inize a use case into classifier/router", "gavinHuang/jevinize → simple-jev not TypeSafe", "featherless-ai/simple-jev", "compare saved decisions / same label can still change the branch", "VihaanAgarwal/jev-diff ≠ Saik0s/diffusiongemma-jev-macos", "not tested with a live Jev API key", "constrained logprob + temp/Platt ≠ Noul", "OpenJevPro pastes openjev-sglang JevBench as own", "zhangcy122/OpenJevPro ≠ IamBusy/OpenJev ≠ ekzhang/openjev-sglang", "PolyForm Noncommercial", "SmolLM-135M / sub-70ms / 0 output tokens", "demo P(True) 0.5052 / Choice conf 0.2872 / Score conf 0.0055", "README claims MIT / GitHub license null / no LICENSE file", "patelvishwa112/jev-system-one-rlcd ≠ arnabgho/rlcd-lite ≠ blackwood-rlcd", "source-backed Awesome Jev radar / 306+ commit-pinned", "logicrw/awesome-jev-projects ≠ AnotiaWang/awesome-jev ≠ yibie/awesome-jev ≠ cobanov/awesome-jev ≠ rupeshpoojary9/awesome-open-system-one", "auto GitHub sync / Issue-only submissions", "hashed n-gram encoder / rival-aware attention", "olanotolu/jevbetter vs jevlike starter", "synthetic hard menus top-1 0.916 vs 0.873 / ECE 0.0182 vs 0.0367 / 40 vs 4608 menus/sec", "shuffled-context control 0.335", "Turn any open LLM into System-One Jev", "uspraveen/Jevify ≠ Mintzs/jevify ≠ gulagala001/jevify", "Jevify-any-LLM architecture probe", "description-only stub / size 0", "Train encoder-only calibrated decision models from a task sentence", "Exu is a toolkit, not a method", "strictly proper scoring rule", "Pre-alpha", "Ruivalim/exu-base", "scratch-trained calibrated decision model", "typed Q → probability dists", "Colvin0315/MiniSystemOne ≠ hyusi2003/MiniSystemOne", "no published weights download URL", "90.5 seconds / 29.2% pipeline evidence", "p_i/p_j independent of other candidates", "Recipe for calibrated decision models — small model out", "init → synth → train → eval → serve", "91.1 % / ECE 0.022 *theirs*", "Jev zero-shot 75.1", "scienthoon/luce", "Put Jev's three headline claims on trial", "0.5B local GPU", "46x speedup / accuracy identical", "ECE 0.624 sentiment catastrophe", "bigger model worse calibration", "RichardoMrMu/jev-mini ≠ yuki-oshio/mini-jev ≠ r-ms/mini-jev", "System-1 decision engine for local LLMs", "structured choices only", "JSON parse of generated text ≠ Noul", "TypefAI JEV / Journal Entry Voucher", "tapsin/jev-local ≠ us/jev-local ≠ Argos1111/jev_local", "Jev 1.13 reward-model eval across 8 benchmark tracks", "40,940 examples / 0 API errors", "RewardBench v1 92.58%", "Precise IF 50.63%", "goya4140/jev-reward-model-evaluation", "Scaffolding in progress", "Jev vs LLM support-ticket routing", "static + live decision bench", "TypeSafe's own published benchmark", "illustrative simulations, not live API calls", "JevBench v1 — smart/cheap/fast/reliable", "I/C/S/K 25% geometric mean", "classifier.dev fast tier 84.8 is Jev behind its own API", "do not re-fold §78 v1.2 board as new", "Laya (421M) 70.1 now on board", "Zero-shot/few-shot LLM routing", "hard budget filter before Jev", "Jev never asked to perform budget arithmetic", "Jev judges the next state, XState enforces transitions", "simulation uses synthetic keyword fixtures", "catalog gravity", "v-modal/awesome-jev-tools", "★339 live REST", "curation is not endorsement", "crawler-maintained directory", "Daily GitHub + npm sweep, human-merged", "RadRebelSam/awesome-jev ≠ AnotiaWang ≠ yibie ≠ cobanov ≠ logicrw ≠ v-modal", "HF peft SPLADE/BGE reranker", "rdxtremity/jev-reranking ≠ carlaiau/jev-reranking", "query-side encoders, not a Jev replica", "ONNX System One Qwen3.5-4B scorer", "source:pngwn/system-one-qwen3.5-4b-scorer", "CC-BY-NC-4.0", "temperature 1.75", "transformers.js AutoModel cannot load this graph", "Consistency benchmark Space", "This Space contains no benchmark result yet", "12-case plumbing fixture", "Benchmark-driven Jev router and judge", "cheap alone is not success", "Jev does not write, sum prices, or claim accuracy %", "Sol 94.2 / Luna 83.9 / Jev path 89.7", "19.2% Sol / 62.3% cost save / 4.5pp miss of 2pp non-inferiority", "p50 latency worse than Sol due to routing overhead", "erendikmenn/jev-llm-router-benchmark ≠ jev-rag-benchmark ≠ ryantsai/jev-llm-router", "Express + node:sqlite", "mock and Jev decision engines", "previous_ticket_count >= 3 is code", "MIN_CONFIDENCE 0.6 still soft", "substring false positives", "aesaganda/jev-ticket-router ≠ SarathChandraBellam/jev-vs-llm-ticket-router", "Universal Figure & Diagram Router", "confidence ≥ 0.85 hard-gate is theater", "generative AI banned from scientific plots", "six visual branches", "hoangngochuong24947-gif/jev-figure-router", "human-labeled (state, question, label)", "166,054 rows / 22 configs", "soft_label for human uncertainty", "Praveenrajus/jev-bench ≠ fstandhartinger/jevbench", "ternary bonsai System One GGUF", "openjev's mechanism, Bonsai's weights", "Hub does not ship weights", "100/100 easy T/F is not Harbor", "label_mass ≠ correctness", "stock llama.cpp Q2_0 silently gibberish", "NicolaiMTLassen/open-bonzi-jev ≠ NicolaiLassen", "transformers.js DeBERTa ONNX", "source:com-kotobalabs/open-jev-deberta-v3-large", "temperature 1.05", "AutoModel from_pretrained works", "onnx-community/open-jev-deberta-v3-large-ONNX ≠ system-one-qwen3.5-4b-scorer-ONNX", "107★ densify", "GH 151M vs README 149.6M", "PR #1 now closed unmerged", "do not re-fold §71 claim-audit as a beat", "typed decisions, RLCD, confidence-gated routing", "structured ≠ correct", "mock not live API", "26 tests", "wjdjdakf17/jev-study ≠ baekenough/jev-study", "bonzi-27b-v2 / ternary-8b / 27b-v1 GGUF family densify", "WANLI-256 74.6% / 65.2% / 71.1% *theirs*", "Bonsai 1 27B Q1_0 runs on stock llama.cpp", "ternary still needs PrismML fork", "hf:heman10x/openJev-verdict-2.0 twin tokenizer-only", "OpenJev Vision image classification + uncertainty", "CLEVR-4 held-out joint 0%", "hfdataset:IamBusy/OpenJev-Vision-Research-v0.1 12,832", "294,912 derived targets not independent samples", "Laya multilingual ONNX WebGPU typed-decisions port", "63/63 selected answers / 5.1e-4 CPU / 1.2e-2 WebGPU", "UpHash-Network/mini-jev is yuki-oshio transfer", "jev-injection-bench 11,900 labelled prompts", "Jev best ranking / Haiku better ECE 0.021 vs 0.058", "0.5–0.9 band is where Jev's numbers do not mean what they say", "Prompt wording moves panic 28%", "manojlds/jev-dspy-bench ≠ dspachos/jev-dspy ≠ jmanhype/jev-dspy-lab", "Jev agreement is similarity, never ground truth", "no aggregate quality grade or merge gate", "AbstentionBench-on-Jev rank 1 of 20 vs 2025 field", "question-asymmetry", "forward-looking 0.465 never extreme", "openkev calibration layer not a runtime", "ECE vs coverage independent", "select_threshold returns inf", "escalation catches uncertainty not ignorance", "misakaikato/openkev ≠ jaredpalmer/kev", "pdf-race Docling→Jev vs Gemini", "parser owns the wall clock", "12/12 tie is a tie", "titles selected not generated", "flopcheck 16 calibrated tweet judgments", "mechanical tells in code", "ZeroX-01/jev-atlas ≠ Zaious/jev-capability-atlas ≠ gorock007/jev-atlas", "Laya calibration lab Gradio MCP", "T never changes argmax", "confidence ≠ top-label p", "easy probe set refused", "40–48 rows too small to ship T", "Gemma-4 26B-A4B jevify classification+calibration", "LoRA adapter twin not independent eval", "Gemma-4 E4B jevify", "E4B LoRA stub card", "kushalpatil/jevify-gemma4 ≠ Mintzs/jevify ≠ gulagala001/jevify ≠ uspraveen/Jevify", "GH kushalpatil07/jevify 404", "PAWS 0.580/ece 0.288 is the weak cell", "smaller E4B slightly better OOD ECE than 26B-A4B", "Hub jevify merged LoRA ships weights", "bonzi Bonsai-8B v1 GGUF densify", "Bonsai-1.7B v1", "Bonsai-4B v1", "WANLI-256 64.5% / 60.2% / 52.0% *theirs*", "rank #4 / #5 / #6 of 6", "JulesHuisman/jev-eval scaffolding / README SHA c356a584 (was empty e69de29b)", "JulesHuisman/jev-eval ≠ SivletLabs/jev-eval ≠ willkelly/jev-evaluation ≠ 4esv ≠ xxkuboxx ≠ onlyoneaman ≠ dayhaysoos/jevals", "7 bands 6/10 vs 40 bands 0/10", "source receipts + confidence slider re-policy without re-inference", "32/32 synthetic is smoke not production", "classify HF datasets across typed semantic dimensions", "roadus2 watch misspelling; lock roadius2/ultra_laya", "ultra_laya REVIEW defects", "default branch claude/laya-jev-review-gg5ppo", "XNLI EN 88.3% ECE 0.032 → RU 77.3% ECE 0.096", "Δ −11.0 pp [−14.2,−7.8]; ECE +0.063", "MASSIVE no detectable difference at n=600", "confidence is function of p_max (r=1.000)", "pointer-not-generator 400 human-authored responses", "proposed ≠ authorized", "FewRel 160: Jev 85.0% vs lexical 13.125%", "gated 100% (95/95) coverage 59.375%", "J++ composable semantic computation language", "judge-jev 0.5 still soft", "947 repos scored; A 273 / B 302 / C 372", "LLM rubric ≠ benches", "No benchmark winner is claimed", "phishing: naive 62.6% vs regex 91.8%; 5-atomic + LR 95.0% *theirs*", "AITuber tension ±15", "README npm global; repo is Rust", "git-confess code owns counting/blame/ratio", "httpx exhibit 11% (13/119) *theirs*", "90d trend +12.40% vs random +12.75% vs BH +41.71%", "5m win rate 25%", "Awesomejev 656 entries / 38,160 stars", "tracker likes 64 (+4) lastModified UNCHANGED", "Laya present; Blackwood ABSENT; Archer still promised_not_landed", "Blackwood tracker ABSENT; likes 2 gated manual", "r = c - p_a", "ECE 0.021; acc 0.807 vs warmup 0.746", "Independent primitive", "11.57s vs 54.10s · 4.67× · 120/128 *theirs*", "default path is pretrained Gemma probs not trained RLCD head", "GH Meanblock 404; lock leesk212/JEV-CPU", "softmax over letter slots ≠ Noul", "WANLI 0.741 vs openjev v2 0.77 *theirs*", "3-way NLI ≠ Noul", "priority 0.464 = majority floor", "banking77 contaminated", "raw margins not probabilities", "do not distill Jev as teacher of record (they distilled Haiku)", "“0.9 is not one number”", "ranking ≠ calibration", "banking77 0.8–0.9 stated 0.86 actual 0.73 over-confident *theirs*", "≠ Praveenrajus/jev-bench ≠ fstandhartinger/jevbench", "$0.0000153–$0.0000226 vs circulating $0.0004 (~20×)", "Score is 0..n-1 expectation not 0–1", "Noul has no confidence field", "TCP floor 198.8 ms", "type reliability is not a reason to choose Jev (json_schema 5/5)", "gateway tax not one number", "Function-only 5/8 vs hybrid 8/8", "4/8 without Jev", "8 designed cases not conversion lift", "200-row pilot Jev 86.5% 173/200 vs Gemini Flash-Lite 86.0% 172/200 vs Pro 87.0% 174/200 *theirs*", "not a ranking", "情緒測謊器", "8-example Jev vs GPT-5.6 Sol ~64× cost 5.4× latency *theirs*", "synthetic; no inference", "≠ JevBench v1.2 §78", "Judged 3317 / listed 2560", "Jev judges, code applies policy", "APA “microsecond policy / zero hallucination” overclaim", "Client-side quiz; pointer from held docs; scanned-PDF warn", "Jev judges / agent reasons / user decides", "selecting an option is not permission to implement", "pattern exact, judgement must clear floor", "no matching pattern → no model call", "not a correctness oracle", "Spec vs artifact remainder", "treating 0.85 as 85% / minProbability hard-gate as Harbor", "VERIFY acquires discriminating evidence, never same-pool confidence-only rescoring", "fast/full/max are ceilings not sizes", "Solar writes, Jev chooses NEXT ACTION", "do not reopen or amend PR #23 or #24 or #25 or #26 or #27", , "Calibration is not alpha", "NO CURRENT ALPHA CANDIDATE", "ΔR² approximately +0.00084", "Brier 0.2131387", "ECE 0.0421875", "Adding Jev probability to deterministic volatility improved Brier by only 1.4058e-05", "default 0.5 keeps zero non pinned", "keepResult median 0.14 to 0.17", "keepCall median 0.28 to 0.35", "usable range is about 0.10 to 0.25", "7.8% to 57.9%", "judges results it never sees", "task-finish eval not built yet", "$0.002 per compaction", "slavadubrov/sgr-judge-bench ≠ slavadubrov/jev-judge-bench", "Jev 108/120 $0.083 0.34 s", "Luna SGR 114/120", "paired Jev accuracy-difference intervals include zero", "not evidence of equivalence", "GLM SGR 26/120 93 format failures", "Terra-planned Jev hybrid 55/120", "rule-based by default, optionally Jev-backed", "empty README", "missing key cannot break the experience", "prefill plus exactly one decode", "softmax over A/B/C ≠ Noul", "BBQ 9,053/10,000 (90.53%)", "ECE 0.0890", "Mean confidence 0.9943", "overconfident", "score and noul not implemented", "DGUI 12 rows (was 6)", "INSTRUCT 119 rows likes 2", "encode the state once, decide everything in parallel", "0.740 accuracy against a 0.508 majority", "ECE 0.047", "fine-tune's advantage ends where its 384-token training data does", "jasonkneen/open-jev ≠ pngwn/open-jev", "same sha d41dc3cd", "Space does not call Jev", "recomputes routing from saved probabilities", "200-case Jev 97.0% / 100.0% / 95.0% / MAE 9.22", "synthetic repository benchmark", "Jev evaluations are advisory", "YehuiTang0316/jev-nlgrep ≠ Bentlybro/jevgrep ≠ can1357/jegrep ≠ uehaj/jev-semgrep", "default threshold 0.8 still soft", "40-line windows cannot prove whole function", "token-native sequential start/end Choice", "Gemini/Haiku stubs not configured yet", "handful of hand-written examples, not a benchmark", "Jev judged exactly what it was given", "laguagu/jev-skills ≠ laguagu/jev-evidence-lab ≠ Pleo2/awesome-jev-agent-skills", "contract_passed is not a claim of guaranteed factual truth", "Wilson lower bound 0.85 floor", "fixture mode no savings claim", "SemIf 2207★ (+21 vs §110 2186)", "jevlike 1043★ (+5 vs 1038)", "TypeAR 15★ (+1 vs 14)", "AnotiaWang 97★ (+1 vs 96)", "yibie/awesome-jev 506★ (+16 vs 490)", "Laya likes 822 (was 802)", "tracker likes 64 flat, lastModified UNCHANGED", "do not reopen or amend PR #23/#24/#25/#26/#27/#28", "Heman10x-NGU/Verdict-open-jev ≠ Heman10x-NGU/openJev-verdict-2.0", "TF-IDF + LogReg ECE 0.0207 vs Jev 0.1440", "Verdict-open-jev 48.07% vs Jev 90.80%", "abstention combined recall 10.00%", "p50 35.58 ms", "K=25 (maximum capacity) 72.00%", "0.85 coverage 84.60% selective risk 1.18%", "26.1× faster than standard Qwen JSON generation", "Jevify 90.0% / 167 ms CUDA graphs disabled", "Finding 1: Brier on stated confidence alone is a trap", "grpo_rlcr 0.78 / ECE 0.084", "reliability 0.007 but resolution 0.000", "27 900 schema-driven decisions", "13 600 / 13 600 questions", "candidate mass min 0.99999624", "22 configs · 166,054 rows · 4 calibration-gold", "sha a39eba3f", "Student B MAE 0.148 / Pearson 0.836 / 86.0%", "pngwn/open-jev-laya-bench README 404", "sha 9f69c742 likes 2", "HDFS 0.9933 (745/750) / retain 0.0084", "BGL ERROR/FATAL protection 1.0000", "2,479 / 2,500 HDFS uncertain", "cache hit 0.9648 (2412/2500)", "$0.153936 estimated", "E2 recomputes from saved probabilities", "Space sha eda59e0a", "MASSIVE English 0.783 / Khmer 0.033 / Hindi 0.133", "40–48 rows too small to ship T", "T never changes argmax", "siren2345/jev-single-decode-transformers ≠ siren2345/jev-single-decode", "Split Transformers experiment from llama.cpp runtime", "tanayvasishtha/jev-lab ≠ dairui1/jev-lab ≠ BrendanH18/jev-lab ≠ yibie/laya-jev-lab", "Four experiments stress-testing TypeSafe's Jev: calibration, bundle bias, label bias, and ensembling", "second pass must be $0.00 from cache", "The pages never call Jev", "Gemma 4 31B 77.0% / Jev 1.13.0 61.4% / Laya 322M 0.0%", "restriction state 95.0% against 84.4%", "None of the systems are particularly good at knowing when to stop and ask", "They skip the question and call a tool directly", "100% schema pass", "six-field joint 48.8% vs 72.8%", "ywchiu/jev_benchmark ≠ Running-Dolphins/jev-bench ≠ Praveenrajus/jev-bench", "ACT / REVIEW / FALLBACK", "A provider failure, timeout, malformed output, or missing answer is **not** a policy outcome", "confidence is descriptive provider output, not a substitute for probability", "Quality denominators include only valid scored answers", "an exact halfway tie chooses the lower level", "aiwithenoch/Jev-Skill ≠ simplosophy/jev-skill ≠ laguagu/jev-skills", "The local path does not claim to turn a smaller checkpoint into Jev", "Low support becomes decision: \"review\"", "MIT-0 SPDX NOASSERTION", "current-llm", "结构兼容，不是 Jev 模型能力", "altryne/jevify ≠ Mintzs/jevify ≠ gulagala001/jevify ≠ uspraveen/Jevify", "Find where Jev belongs. Design the questions. Measure the difference", "TypeAR-AI/TypeAR 301 → TypeLLM/TypeLLM", "TypeLLM/TypeLLM 16★", "SemIf 2241★ (+34 vs §111 2207)", "jevlike 1051★ (+8 vs 1043)", "AnotiaWang 98★ (+1 vs 97)", "yibie/awesome-jev 525★ (+19 vs 506)", "Laya likes 864 (was 822)", "tracker likes 67 (+3 vs 64)", "lastModified UNCHANGED `2026-09-20T04:29:16.000Z`", "do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#32", "hysteresis enter/exit / replay policy without inference", "calibration does not compose / hop-ECE permutation-invariant", "equal-width vs quantile ECE / ranking ≠ calibration", "Qwen2.5 ≠ Archer / Qwen 3.8 sparring ≠ Archer / Qwen/Qwen3.8-27B ≠ Archer", "Deferred Crispification / TCE / AMS", "g0runmezadam/what-is-jev IS tunahansahin897/what-is-jev", "pd.cut equal-width vs jeval quantile", "A hunch is a probability with a policy attached", "soundness theater / measurement theater / hourly 0843", , "Jev Capability Resolver / NiazMorshed2007/jcr", "one tool nested capability tree / returns context / does not execute", "skills vs capabilities / workflow+judgment vs operations", "format independent of Jev / proposed open standard", "JCR_BAND_RATIO 0.6 is application policy / soft scores ≠ hard gates", "routing ≠ permission / docs ≠ authority to run", "sol-vs-opus5-20 lookup+explain / n=1 / Not Harbor task-execution", "wall-time mixed / Sol slower with JCR in 19/20", "NiazMorshed2007/jcr ≠ skill-broker ≠ skillranker ≠ jev-sift ≠ jev-lens ≠ jevusher ≠ jev_select_capability", "do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34", "notes.md §116", "copy the SemIf/MLX installer?", "quote 5.21× as beating Jev?", "treat 0.845 as a TypeSafe replica?", "collapse SemIf into kw2828/zhihz/semif-rs/semif-serve", "softmax over options as a Noul", "llm prompt to jev primitives", "conversion assistant not equivalent behavior", "heuristic conversion ≠ calibrated Noul", "alexwestco/llm-to-jev ≠ altryne/jevify", "user-provided 0940 / notes.md §118", "judge ≠ actuator", "candidate_mass", "softmax over A–H ≠ Noul", "hourly 0947 / notes.md §119", "ggmlc GGUF is not llama.cpp", "serving substrate ≠ calibrated replica", "Qwen3.5-9B ≠ Archer", "planner writes JEV selects", "hourly 1049 / notes.md §120", "open recreation ≠ calibrated replica", "semantic lint is a sensor not a proof", "cutoff 0.8 still soft", "paired bootstrap CIs *theirs*", "Same accuracy, 35x faster *theirs*", "hourly 1143 / notes.md §121", "revisit HIGH / since-last-look", "catalogued repo changed", "star-noise vs material change", "densify prior notes without inventing equivalence", "decide is not generate", "tryDecide returns typed calibrated judgments not a token stream", "GLiNER/GLiClass ports are class members not Jev replicas", "93.5% *theirs* not Harbor", "74.9 *theirs* not Harbor", "8.7x *theirs* not Harbor", "Option-Marker joint attention", "openjev:0.2.1", "thinking=True/False per-field budget", "PLAN_Qwen35", "hyperspaceai/jevcache ≠ kushals256/jevcache", "wire-compat ≠ logit-equiv", "SHA move is not a replica", "hourly 1248 / notes.md §123", "typesafe-sdk 0.7 Pydantic response models", "msgspec dropped", "The server's output is unchanged and was never wrong", "SchemaError is 400 plain-string detail not 422 list", "Pydantic response models ≠ logit-equiv", "msgspec dropped is not a replica", "Error contract is not a Noul", "coverage-at-error-budget *theirs* not Harbor", "PLAN_Qwen35 still proposal for review", "GLiNER locate ports are class members not Jev replicas", "Locate ≠ decide", "~160 ms *theirs* not Harbor", "0.971 F1 *theirs* not Harbor", "hf:fr0stbit3/laya-gguf serving substrate ≠ calibrated replica", "jkcdarunday/SystemOne-Next ≠ TypeSafe System One", "hourly 1340 / notes.md §124", "vLLM NVIDIA + MLX Apple Silicon", "Codiv hosted free endpoint", "dual /v1/systemone + /v1/chat/completions", "chat 501 on MLX", "dual serving is not generate", "Hosted Codiv ≠ TypeSafe", "hr98w/jev-visual 167★ Apple Silicon visual candidate scoring", "37.30s → 2.40s at 64 decisions *theirs*", "Breakout 9 bricks 6 returns 2 lives *theirs*", "candidate probabilities are relative not correctness", "jkudish/jev-mcp 156★ ten MCP tools", "recommendation is advisory", "the server never blocks on its own", "TypeSafe CLERC 5% to 18% *theirs*", "jkudish/jev-mcp ≠ burnigtm/jev-mcp", "zhengxuyu/litjev off-the-shelf Qwen decision layer", "Probabilities are not calibrated by default", "Qwen/Qwen3.8-27B ≠ Archer", "zhengxuyu/litjev ≠ alexwestco/llm-to-jev", "Zefan-Cai/Open-Jev LoRA + scalar head", "2B 94.71% 9B 97.54% hard test *theirs*", "2B OOD 86.02% 9B OOD 91.97% *theirs*", "80,816 training rows", "27B still in progress", "LoRA ≠ RLCD replica", "Zefan-Cai/Open-Jev ≠ TheoLeeCJ/openjev ≠ razorback16/openjev", "cristianoliveira/jeq intelligence you can pipe", "pass-min 0.8 still soft", "JEQ does not own actions", "AndyInQtr/laya-coreai CoreML serving substrate ≠ calibrated replica", "AndyInQtr/laya-coreai ≠ mizorewww/laya-coreml", "hourly 1441 / notes.md §125", "TypeLLM/TypeLLM densify HEAD 6a48f9f1e623", "README densify 3k→12k B", "Batch 5.8x *theirs*", "Constrained AR ≠ calibrated Noul", "jaredpalmer/kev densify HEAD b339f446a0ef", "Kev-0.6B 4B 8B family", "4B new-source 0.790/0.806 *theirs*", "8B new-source 0.796/0.780 *theirs*", "Jev hosted 0.857 *theirs*", "Questions share the input text but cannot read each other", "No Jev outputs were used for training", "8.2% ≥0.9 on wrong *theirs*", "option order can change an answer", "Qwen3 ≠ Archer", "TheoOliveira/pi-jev 21★ fail-closed routing", "JEV_THRESHOLD 0.65 still soft", "harshwasan/jev-sentinel fail closed never auto-allows", "harshwasan/jev-sentinel ≠ leepokai/jev-guard", "jackbarunz/jev-tool-router ≠ esinocchi/jev-tool-router", "threshold 0.90 still soft", "76/81 vs 77/81 *theirs*", "0.419s vs 2.459s *theirs*", "$0.00486 vs $0.03673 *theirs*", "not a security boundary", "baronunread/leanest fail-open uncertainty means RUN", "classifier.dev default Jev/Laya pluggable", "openlayer-ai/jevals ≠ dayhaysoos/jevals", "estimates not Harbor", "classifier ≠ authorizer", "MrJev/awesome-jev 118 entries catalog ≠ endorsement", "MrJev/awesome-jev ≠ yibie/awesome-jev", "Koushik890/jev-firewall fail closed ask_below 0.7 still soft", "CompleteTech-LLC-AI-Research/jev-codex-approval experimental native not compiled", "confidence is not a measured probability", "rh-guard owns primary gates", "hf:rAVEUK/open-jev-deberta-v3-large encoder class member not Jev replica", "hf:p-yan/laya-quanto serving substrate ≠ calibrated replica", "hf:Gtrkrsk/laya serving substrate ≠ calibrated replica", "hourly 1542 / notes.md §126", "razorback16/openjev densify HEAD febf02e88989", "release 0.3.0", "re-pin vLLM PR #57250 restructured head", "MODEL_VERSION stays openjev-0.1", "uv.lock hygiene", "restructured vLLM head ≠ logit-equiv", "frostney/clean-code-review 7★ typed judgments not opinions", "documentation is read not judged", "morcoan/JMP Joint Model Participation", "Models participate. Real tools execute.", "Thresholds are policy not model", "Kelbie/hunch ≠ carldaws/hunch ≠ tpellet/hunch ≠ huncho", "Jev never generates prose JSX or code", "json-render is the only renderer", "game success ≠ calibrated Noul", "Shalimov04/open-jev ≠ razorback16/openjev", "MstyAI/laya-onnx empty repo", "hf:Praveenrajus/jev-bench HTTP 200 was 401", "hourly 1643 / notes.md §127", "TypeLLM/TypeLLM densify HEAD 702e6a287f3c", "truncated thinking then constrained decode", "0.8B thinking On 0/18 *theirs*", "forced closure 20/20 type-valid *theirs*", "jaredpalmer/kev densify live HEAD 8465c4c4c294", "Kev-0.8B completes family", "4B new-source 0.794/0.832 *theirs*", "9B new-source 0.812/0.837 *theirs*", "transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*", "SemIf Kev-9B 0.917 Jev 0.965 *theirs*", "scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*", "transformers >= 5.17", "Qwen3.5 ≠ Archer", "notque/vexjoy-agent 421★ /d routes /do fallback", "Facts go to code. Judgments go to Jev. Only facts can block.", "Jev never blocks", "jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev", "five-lines threshold 0.80 still soft", "371ms $0.0000189 300-call *theirs*", "tpellet/jevify ≠ altryne/jevify", "seb4ez/jevguard-mcp ≠ seb4ez/jevguard", "resumocast/jev-mcp ≠ jkudish/jev-mcp", "Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort", "MidasMulli/kev-ane 155/155 argmax *theirs*", "hourly 1746 / notes.md §128", or "cascade sign-flip / calibration theater": read `references/faq.md`, + other", "training confronts Choice other / none-of-the-above", "soft AGENTS.md rules vs the linter", "screenshot Choice / omni System One", "extractive quotes / pointer not generator", "compaction summarize vs pointer", "encoder vs Jev compaction backend", "shadow-mode compaction rollout", "CI flaky-vs-real merge gate", "fail-open VOI wake/resume", "claim vs session evidence", "S1 indexer escalate-S2", "Harbor on/off routing", "fail-open vs fail-closed wake vs CI gate", "encoder vs Jev computer-use backend", "hybrid local decide + remote fill", "DONE vs verified success", "stdout prune vs session compaction", "OpenCode jev-pruner vs Claude jev-pruner", "zen-chat vs jev-zen Noul", "hard envelope then Noul prune", "Cua-S1 vs TypeSafe Jev", "plan vs execute dry-run", "specialist computer-use vs general agent", "local drop-in vs stub scorer", "route vs memory", "when does it hold / extractable from state", "decision model vs constrained LLM", "dual-process S1/S2", "combinatorial grid vs extractive", "uncalibrated local likelihoods", "decision-native RAG", "classify-first / read selectively", "living applied-mappings atlas / class patterns", "silence as safer / draft-gate heartbeat", "robotics text-state vs pixels", "verbatim ledger vs summary", "judgment as language primitive", "Stagehand extract pick-and-copy", "harness observe-score-act vs demo loop", "public judgment wall / six parallel questions", "meaning-search without embeddings", "attention ≠ correctness", "skills→oxlint / AST prove ∩ remainder", "session-sticky first-prompt routing", "measured RAG rerank vs generative rerank", "capability kernel / secrets never in the agent", "Jev is SENSOR not policy", "type-safe ≠ correct", "typed control plane around DSPy", "native vs verbalized confidence", "engine owns truth / Jev owns judgment", "human-confirmed kill gate", "train specialist vs few-shot hosted", "decide→policy→LLM leftover", "Noul 0.5 cannot-tell never rounded", "calibration ≠ sortable / ORDER BY", "pairwise inversion / Score ordinality / two-decimal ties", "wire-compat GLiFormer /v1/systemone", "class-backend economics", "loopback gateway hosted + local", "do not distill Jev as teacher", "active-learning triage", "evidence-packet explorer", "meaning-grep AND/OR/NOT", "closed-vote-only / no planner LLM", "Jev vs PCD Harbor", "PCD O(1) ≠ Noul", "host-owned handlers × System One", "OMP/pi fail-open gate", "permission vs probability / operator owns thresholds", "judgment ≠ permission / Jev never grants access", "eval integrity / instrument not score", "constrained optimizer + S1 features / never sole hot-path gate", "privilege ≠ verdict / effect contracts not tokens", "attention filter / VOI for human review / never blocks / never green unless sure", "measurement owns endorsement / evidence-gated question packs", "Jev supplies evidence / code owns authority", "ranking ≠ calibration / never hard-threshold raw p as frequency", "hot-click CU / indexed element table", "Jev judges relevance / code decides structure", "local rules first then remainder / never auto-train on own hides", "combinators / System One as control plane", "receipts not leaderboard / type-safe ≠ correct jaggedness", "VOI over skill library / skillranker abstention", "OOD calibration / AUC ≠ ECE", "Jev vs thinking-budget small models", "turnstile / replayable evidence≠authority", "MLX one-pass schema→JSON / Apple Silicon replica economics", "memory leases ended by new evidence", "never confidently wrong / TLA+ compose / escalate instead of hard-gate", "no seal no advance / coverage ledger / mint ≠ product brain", "skill-broker sibling / judgment ≠ permission", "sureness bands / max_prob is generous", "JevBench / calibration not in Main Score", "CI typed gate before expensive review", "Codex MCP host adapter", "judgment as attention redirect / jev-preflight", "compress-before-first-send / dizk jev-lens", "tools≠use / SessionStart over hoping", "observational memory / pi-om keep-kind", "open-Jev class / openvons / JevPick", "physical-world System One / HA-Jev / not for locks", "judgment outside the store / jevql", "landed-script trust / headless≠auto-approve", "digital-design combinators / extended five", "VOI cache admission / same-intent skip LLM", "BM25 vs Jev skill routing Harbor harness", "zeroshot vs BERT / contamination DiD", "typed escalate continue abort baton / inverted loop", "worth-your-attention VOI / ThinkyMiner Winnow", "Jev WHETHER Python HOW LLM WHAT", "conflict vs ignorance / named Choice escape", "Playwright executes Jev chooses", "OpenJev /v1/decide not drop-in", "SemIf wire-compat runoff; SemIf rename densify / MLX backend / 5.21× systems≠semantic / Softmax ≠ Noul (`notes.md` §117)", "decision-as-memory flywheel", "record/replay CI / jevassert", "failure-finding arena / jevarena ≠ jev-arena", "BBQ not a bias cert", "decider≠executor", "sentence-as-rule lint / jevlint", "sentence-as-rule lint / jev-lint is jevlint rename", "VOI hunk prune", "whole-repo intent VERIFIED/VIOLATION/UNKNOWN", "GLiNER2 spec ≠ replica", "open replica substrates / grande / laya-jolt / JEV-CPU", "ONNX local-jev not equivalent", "persist constraints across compaction / pi-heed", "calibration+cost first-class gates", "Harbor-shaped Jev vs SGR LLM-as-judge / jev-judge-bench ≠ jevarena ≠ jevbench", "hand no-text steps / jev-use / Vercel drops confidence", "Pi System-One control plane / pi-jev-control", "never free-generates / jev-gpt tree of Choices", "OpenRouter recipe atlas / samples not benches", "personal history feed / jevfeed / no social graph", "competing NAR claims / dual-channel ECE / openJev-verdict ≠ OpenJev", "empty compaction-proxy skip / IPECTER", "throughput ≠ latency / like-for-like ECE", "1-token logprob endpoint ≠ Noul / coverage ≠ correctness", "open replica engine / jevinf", "unofficial Elixir SDK ≠ OTP peer", "jevex n=16 files-to-read VOI", "commit pre-review attention≠verdict / middle band", "Hermes plugin is Agnes not TypeSafe", "pi-jev-compact ≠ pi-jev-compaction", "empty Codex-proxy skip / IPECTER runway", "decision-native inbox / mailordinal", "unofficial jev-cli not ready / ≠ jevql", "laya-multilingual / English checkpoint confident-wrong OOD", "schema-scorer peaked ranking ≠ calibration", "HF 401 / GitHub 404 Hub-only", "productized System One HTTP / classifier.dev", "escalate-under-threshold / smart tier / multi-label ignores", "silent FALLBACK / granite 0.546 vs advertised 0.800", "vs_jev tracked JSON / read eval/README", "choxos/jev-reviewer ≠ egma-ai / systematic-review pointer", "two-pass Choice+Noul evidence extraction", "not-found is an answer", "human check as productized judgment", "githubnext/localjev ≠ kunchenguid/local-jev", "wire-compat ≠ logit-equiv / prompted JSON ≠ structured read", "self-reported probs / entropy confidence", "GitHub Next local /v1/systemone", "LM Studio runner gap / structured-read primitives", "NandhaKishorM/laya packaging ≠ Hub-only / Router script-before-p", "post-T ECE ≠ raw ECE / Banking77 token-budget", "0.85 still soft / not TypeSafe drop-in", "external census ≠ scored bake-off", "GLiNER2+routers class-boundary", "incomplete openjev census vs watch", "Harbor honesty watch / silent fallback", "JevBench v1.2 geometric mean / cal ON rank / weight sensitivity", "option-order 72→21 / instruction models in the class table", "self-host latency ×2 assumption / est. costs", "Laya absent is a gap not a named exclusion", "Qwen3.8 27B ≠ Archer", "hourly already-folded watch / apply-the-five / skip thin noise", "hard-gate Noul as PR/quality gate is soundness theater", "S1 never stalls waiting / S2 one-use advisory", "Local controller ≠ githubnext/localjev", "purple telemetry = consumed not arrived", "seed = geometry not async replay", "20% starting gate still soft", "no pixels to either provider", "OCR+AX observe-score-act / typesafe-computer-use", "never send screenshot to frontier for the decision", "overlapping CU options = false low confidence", "split kind/item/site", "155× one-screenshot ≠ Harbor taskset", "decision ≠ answer-reader capture", "ASR observe-score-act / jev-voice-browser", "partial-speech VOI / free-text waits", "spoken confirm ≠ hard auth", "numbered overlay without another model", "wrap-as-execution / AgentGhost ALLOW ASK DENY", "rules first then Jev remainder / ASK throws / fail-closed", "reddpy/AgentGhost ≠ jwen5419807/agentghost ≠ vventirozos", "JP genre atlas / studio_yebisu / stars ephemeral ≠ eval", "Jev Clearly Explained / akshay_pachaar / LLM hammer", "schema-safe ≠ correct / 200× 400× TypeSafe ceiling", "questions-as-code / shadow first / not a TypeSafe how-to", "proposition ≠ embedding / contrast-set", "boolean composition of soft Nouls / AND OR NOT", "uehaj/jev-semgrep ≠ semgrep.dev", "meaning-grep dedicated fold / not a gate", "decision-validated UI / Jev never authors text", "decision-as-assert / jevtest ambiguous band", "typed decisions drive UI / jev2ui", "hybrid S1 closed verb menu / anima3", "pointer-not-generator search / JevFind", "jev-frontier-bench ≠ frontier-100", "product bakeoff ≠ architecture duel / GLiClass", "four engines same questions / majority floor", "authorship named escape / not evidence", "ha-switchboard HA remains execution", "n8n classify/route/score / Low Confidence", "fast-jev-compaction-pi ≠ pi-jev-compact ≠ pi-jev-compaction", "jevloop full-distribution optimizer / no LLM in the loop", "laya-vision SmolVLM / score untrained", "Cerebellum-2B /v1/decide ≠ TypeSafe / wire-compat vs agent-routing", "laya-grounded not drop-in / Platt not temperature", "GestaltLabs/Jeff-1 ≠ logan-markewich/jeff / acc vs ECE n=9730", "stanley-code empty findings ≠ approval / human promote", "findme ≠ JevFind / NL memory beam-search FS", "jevsubrouter price workers not conversation / counts ≠ dollars", "feelings .feels() default 0.5 is Noul-0.5-never-rounded / ≠ hunch ≠ Probably", "apa-agent-harness ≠ AntonioCoppe/jev-harness / unpublished npm", "grok-bot-jev skill cannot force a bot that ignores it / A/B proxies not tokens", "Essentiel-Jev never authority / human every action", "enzo-mcp independently falsifiable claims / ≠ jev-sift", "pigeonhole OTHER skip / decision-as-filing", "jev-reliability Nothing about accuracy", "clduab11/jev-test ≠ realZachi/jevtest / Nothing runs yet", "jev-rag-benchmark Jev wins is not an assumption", "dairui1/jev-lab ≠ BrendanH18/jev-lab", "jevmail gmail.readonly / mailjay archive/trash", "ZHUBoer/ego-jev reserved __none__", "runWorkflow completed ≠ success", "jsort scores are relative", "Noul not Choice for scale", "groundedness-judge-bench native vs schema-guided", "implicit_true included in yes", "jev_playground 0 promotions", "routing-backtest 0.0447%", "yuyang2230/jev-agent-skill jev-1.13-free", "jev-techstack-classifier stack_config.json", "s1_ruby collapse late", "undecided? abstain", "2389-research/judgement license null", "confidence ≠ winner p", "typesafeai-sdk-community not a new species", "tpellet/hunch exit 3", "never-execute list", "jev-file-search scores not calibrated accuracy", "jev-linkmap Jev never sees S2 prose", "muhammedilyasy/jev-mail metadata only", "tidy none-of-folders stay", "tab-bouncer pinned/audio/current never closed", "lkclean Show fail-open", "jev-yt-time-saver Show anyway", "ORIGIN pause-if-no-Jev", "validResponse sums-to-1", "jev-crawlers risk bands never raw boolean", "jevbrain AUTO_ACT is not a Noul", "judgekit YAML classify/score/route/verify", "typed-judge-kit verdict-in-code", "alsoleg89/decide packing VOI", "0.8 ≠ 80% accuracy", "Jev-Calibration Platt ECE 0.117→0.052", "jev-calibration-arena never acts", "ctmx/openrouter-jev-mcp Decision-as-Plugin", "FrancoisChastel/jev-code ≠ npm jev-code", "claudecode-jev-marketplace fail-open not hot path", "pedroknigge/mcp_jev packs not ask_jev", "cyrusasco/typesafe-mcp noul deadband 0.35–0.65", "codaaiteam/jev-skill jevtypesafeai.com ≠ TypeSafe", "hermes-switchyard ≠ hermes-jev-router ≠ hermes-plugin-jev", "nanoprune 2.8MB ECE 2.58%", "smartdio/jev-browser-agent ≠ ZHUBoer/ego-jev", "Dakai/omp-jev-web DONE ≠ proof", "hari007sh/jev ≠ dannote/jev", "0thernet/system-one-skills deterministic verify", "typed-gate band [0.40,0.60] is refusal", "pi-jev-gate fail-closed; choice is the verdict", "Foq ~25ms/2.2GB local", "rev prefill-only + HF jev-0.5b", "robfrase/jev planning memo", "typesafe_agent_gates 27/27 / 31/31", "EpicEric/safe-sh static remainder", "pastepilot Confirm before act", "Jev-Reranker live Jev not yet measured", "sessionwise opt-in relevance", "jev-search pointer sieve", "400ms Salesforce WebMCP", "typesafe-scheduler-diagnostics advisory", "droidjev screenshot-free", "Tewoto1 jevcu planner still writes", "ha-conversation-jev Jev→Grok", "dsh-jev can only gate", "jev-classification-benchmark specified not run", "jev-luna-pagerduty p≥0.50", "meldltd/meldecision laya-go ONNX", "laya-doom never pixels", "logixism/laya-api empty README", "akpsahan/laya ≠ Archer", "choxos/jevchess engine owns truth", "jev-drive sim not AV", "story-arc Jev never authors", "jev-hs-assistant HS6", "golergka/jev-plays-starcraft-2 UI-verified ≠ API Victory", "awesome-jev-use-cases catalog", "Nibir1/typesafe-go ≠ official", "fingerprint after redact", "recall vs decide", "publish fingerprints+answers", "CI replay as Harbor cousin", "Cache hit ≠ correctness", "hyperspaceai/jevcache ≠ kushals256/jevcache", "human labels only", "score never auto-accepts", "production capture flywheel", "sutro-sh/jev-align ≠ caiovicentino/jev-align", "guidance ≠ hook", "catalysts ≠ summaries", "compile-time System One", "unofficial ≠ TypeSafe", "format_version modernbert-jev/1", "Argos1111/jev_local ≠ us/jev-local ≠ kunchenguid/local-jev", "LFM default ≠ ModernBERT backend", "Nemotron ≠ TypeSafe Jev", "not a calibrated replacement", "djev-dev complements djev-spark", "images as Choice options", "Laya essay numbers *theirs*", "Router/OOD confidence", "hosted bootstrap ≠ silent TypeSafe", "difficulty + policy thresholds + JSONL trace", "jev-codex-pilot model + reasoning depth", "keep/shadow/hybrid/reject", "quarry evidence projection", "Frank-ZY-Dou/awesome-jev robotics/3D/control", "one-dollar-tahoe TypeSafe Jev defense eval", "jevguard calibrator/cache/escape", "jev-ci-selector CI shadow mode", "llama-jev llama.cpp replica", "petercr/jev-orchestrator ≠ FleeexCorp/jev-orchestrator", "seb4ez/jevguard ≠ AseemPrasad/JevGuard ≠ pablozr/JevGuard", "webNeat/llama-jev ≠ WiktorB2004/llama-index-jev", "OpenCode jev-pruner context sieve", "observe→score-candidates→prune", "jev-zen / jev-1.13-free", "zen-chat ≠ Noul", "fail-open original", "keepScore >0.1 floor", "host port of tamaratran/jev-pruner", "indiejoseph/opencode-jev-pruner ≠ nrdz-labs/fast-jev-opencode", "jev-webagent-bench empty stub", "Kiln-AI/jev_jsonschema noul_threshold 0.5", "NSStudent/JevSwiftSDK unofficial", "GLiNER2 native Apple path", "unofficial Swift/Core ML GLiNER 2.5-small", "entity spans + confidence", "not Choice/Score/Noul", "not TypeSafe", "label descriptions as schema", "on-device ANE economics", "honesty locks", "shershah1024/gliner-native-runtime ≠ Fastino", "≠ gliner25-compaction ≠ gliner2-ultrafast ≠ Eran-BA/Jev_from_GLiNER2 ≠ NSStudent/JevSwiftSDK ≠ jevmlx", "default threshold 0.1 still soft", "soft Noul ≠ hard safety", "Decision Graph Protocol frame→assess→commit", "app retains permissions/effects", "Jev-first assessor-neutral", "guarded commit / receipt/next frame", "assessment batching", "hard-gating DGP as safety theater", "numerous-com/dgp ≠ TypeSafe official", "jegrep calibrated path+range Nouls", "no embeddings/index/daemon", "~$0.01–0.03 typical", "agent --json", "can1357/jegrep ≠ Bentlybro/jevgrep ≠ uehaj/jev-semgrep", "Archer-arch fidelity", "kev family OOD 0.76–0.77 vs Jev 0.86", "block-causal isolation", "pointer/readout CE-trained", "/v1/systemone drop-in", "replica honesty", "cost-sensitive decision theory × System One probabilities → control flow", "thresholds derived from costs not hard-coded", "YES / NO / UNSURE from cost_false_yes / cost_false_no / cost_human", "auto-batching same-object questions", "Kungie/gut ≠ tpellet/hunch ≠ carldaws/hunch", "judgment vs generation", "deterministic execution after probabilistic judgment", "exactly one app-owned callback", "explicit uncertain branch", "Illusion47586/judge ≠ lexingtonhibiki/judgekit ≠ Ascurse/typed-judge-kit", "variable-N option scoring as the trainable object", "dynamic candidate bags not fixed label sets", "zwliJay/jev-forge ≠ NanoJev", "open replica economics / latency vs closed Jev", "NAR local drop-in", "wfzyx/von late-catch HIGH", "competing NAR claims / replica honesty", "typed judgments vs chat judges on guardrailing", "ishaannk/llm-vs-jev cross-note only", "deeper integrity fold is rh-guard", "nothing wins outright", "can be argued out of guarding"", "Jev IS the if-statement", "judgments/probabilities drive branches", "text model only writes prose", "interpreter owns variables/loops/budgets/replay", "otherwise maybe / confidence gate", "chaos samples after the gate", "southpolesteve/probably ≠ carldaws/hunch ≠ feelings ≠ Kungie/gut ≠ Illusion47586/judge ≠ tidymodels/probably", "133★ / forks 10 live", "build calibrated classifiers from human feedback", "retrieve by relevance not resemblance", "one calibrated yes/no per memory in one request", "pointer mode 17/18 19/20 *theirs*", "embedding resemblance misses the allergy", "samdotmak/jev-recall ≠ jev-search ≠ jev-sift ≠ carryforward ≠ chopratejas/invalidate", "memory leases ended by new evidence", "six Nouls then fixed rules in code", "0 of 157 false invalidations", "questions/plans/directives are not evidence", "unsure → review queue", "host keeps the store", "name↔body / comment truth / test-claims", "mizchi/jev-lint is mizchi/jevlint rename", "no shipped rule has severity error", "~1 in 5 findings wrong *theirs*", "mizchi/jev-lint ≠ huntedman/JevLint ≠ MichitoSugawara/jev-lint", "JSON Schema → typed JSON via Jev", "noul_threshold 0.5 decoder not a proof", "IncompatibleSchemaError lists every bad property", "on-device Laya CoreML ANE", "~5 ms P50 short decisions", "189/189 FP16 checkpoint parity", "10× not achieved", "mizorewww/laya-coreml ≠ gliner-native-runtime ≠ jevmlx ≠ NandhaKishorM/laya", "softmax over allowed tokens ≠ Noul", "question-first cache", "Micha0827/snapjudge ≠ githubnext/localjev ≠ jevmlx ≠ cendress/SnapJudge", "Jev-first Pi agent loop", "slow-LLM fallback", "explicit action menu / CandidateSource unimplemented", "62 tests wiring not quality", "direwolfiy/JevPi ≠ standardagents/jevpilot ≠ pi-jev-control", "resume-screening bias audit methodology", "name×resume factorial independent Nouls", "callback determined by resume quality", "mean-probability name gaps operationally negligible", "natemoo-re/bias-bench ≠ BBQ", "Plan/PRD panel → code-owned pass|review|block", "cheerleading out of scope", "austindixson/planalyzer ≠ single-goodness Noul", "cost-aware multi-model routing/escalation", "decide vs do", "successful-task cost", "cannacre8ive/switchboard-ai ≠ ha-switchboard ≠ hermes-switchyard", "frozen-protocol zero-shot bench", "TypeSafe Jev vs PrismNLI vs Laya", "contamination caveat", "elcronos/jev-vs-open-decision-models ≠ JevBench ≠ DMB", "context-window admission control", "VOI gate which tokens are worth the expensive model", "fail polarity per lens", "on small inputs lenses lose money", "cvsgireesh/jevusher ≠ jev-sift ≠ winnow", "typed decision control plane", "receipt ≠ authorization", "historical-v0 zero retained cases", "MokiMeow/jev-fabric ≠ jev-forge ≠ dgp", "live 15-dim typed rubric re-score per pause", "scoring economics exemplar", "OpenJev/Codiv ≠ TypeSafe hosted", "jose-troche/live-rubric ~$0.000004 desc / ~$0.000006 README", "adversarial pre-registered Jev eval", "28 predictions before data", "123,805 requests", "confidence does not track ignorance", "polite injection 65% / crude 0%", "willkelly/jev-evaluation ≠ jevals ≠ jev-baselines-eval", "provider-neutral Elixir/BEAM Noul/Choice/Score SDK", "class infrastructure", "nshkrdotcom/system_one_sdk ≠ typesafe_sdk ≠ dannote/jev", "question-linting of Jev questions themselves", "nine jaggedness rules, no API key, no labelled data", "static lint ≠ measured separation", "yodablocks/jevq ≠ tenbin ≠ JevLint ≠ commitjev", "open-weights Laya as class exemplar (binding)", "Nx/Bumblebee runtime", "host chooses backend", "ChristianAlexander/laya_ex ≠ system_one_sdk ≠ dannote/jev ≠ NandhaKishorM/laya", "on-chain/edge Laya deploy", "parity_verified stays false", "model output never grants Tx", "humandebri/IC-Laya ≠ laya_ex", "auditable weekend replica", "Jev outputs never used for training", "soft human-vote distributions", "unpaired 0.577 vs 0.727", "agilabs-ai/jev48 ≠ JevBench ≠ Mapika/decider", "adversarial dual-judge / framing attack surface", "comparative framing is the usable judgment", "prior injection crowds out evidence", "copyleftdev/ember ≠ ember.js", "Laya specialist fine-tune pipeline", "training still GPU-pending", "PIXELZX0/XERON ≠ convaiinnovations/laya", "Hub Laya replica drop", "daliborsb/laya ≠ convaiinnovations/laya ≠ NandhaKishorM/laya", "System One student distillation corpus", "gold is programmatic", "teacher is closed-API clone", "do not distill Jev as teacher of record", "MagaBitmex/jev-4b-distill-data ≠ missing student checkpoint", "non-LLM VIN System One", "planning depth not chat", "lewislululu/jevon ≠ douglance/jevon", "source-bound evidence checks", "local quote mismatch needs no API", "exit 0 ≠ claim truth", "WaynezProg/jev-kit ≠ jonathanavis96/jev-kit (Airlock) ≠ jev-use ≠ jev-mcp", "independent System One evidence catalog", "scores not one leaderboard", "no external record currently reproduced", "TokenTrim no-Jev matched hybrid 62.4%", "reachjalil/system-one-bench ≠ mallahyari/system-one-benchmark", "21 tasks · 134 items · 208 questions", "scenes from public GitHub contracts, not production logs", "SivletLabs/jev-eval ≠ willkelly/jev-evaluation ≠ 4esv/jev-eval ≠ xxkuboxx/jev-eval ≠ onlyoneaman/jev-eval ≠ dayhaysoos/jevals", "option isolation (sibling-blind)", "permutation-equivariant", "Hub OWNER not published", "nafisazizir/hev ≠ jaredpalmer/kev", "frozen local LLM logits, no trained decision head", "residual-head 9,222-param decreased 73/96→67/96", "confidence = 1−normalized entropy, not P(correct)", "yuki-oshio/mini-jev ≠ r-ms/mini-jev", "Jev classifier as autoregressive next-token predictor", "ChatJev-style soundness theater", "erik-dunteman/ChatJev ≠ dannote/jev ≠ jev-gpt", "calibrated decision head × AlphaProof value head", "implementation-layer isomorphism, semantic difference", "timeout = censoring", "do not launder Noul as proof", "parallel rank-prediction vs serial selection", "independent questions can conflict", "zzzzzec/jevsort ≠ keltokhy/jsort", "curated open System One ecosystem catalog", "rupeshpoojary9/awesome-open-system-one ≠ AnotiaWang/awesome-jev", "arXiv paper radar with Jev relevance scoring", "ranking ≠ calibration / 0.5 still soft", "fail-open failed evals not marked seen", "train calibrated ~27M from scratch", "typed Q→prob dist / one forward pass / no LLM decode", "hyusi2003/MiniSystemOne ≠ Colvin0315/MiniSystemOne", "description-only stub / size 5", "ESCI hard probe fails four of six", "jev_bool ECE 0.242 inversion 0.255", "do not re-fold §60 six-gates as new", "jobbyjev one-request-per-company from batch-size result", "find/design/evaluate TypeSafe Jev decision loops", "karanb192/jev-architect ≠ samtay32/jev-system-architect", "Jairik/jev-distiller size 1", "distill-Jev UI stub / do not distill Jev as teacher of record", "post-launch scored use-case map / Jev self-scores then human curation", "licensedsaucer9-web/jev-opportunities", "Jev-inize a use case into classifier/router", "gavinHuang/jevinize → simple-jev not TypeSafe", "featherless-ai/simple-jev", "compare saved decisions / same label can still change the branch", "VihaanAgarwal/jev-diff ≠ Saik0s/diffusiongemma-jev-macos", "not tested with a live Jev API key", "constrained logprob + temp/Platt ≠ Noul", "OpenJevPro pastes openjev-sglang JevBench as own", "zhangcy122/OpenJevPro ≠ IamBusy/OpenJev ≠ ekzhang/openjev-sglang", "PolyForm Noncommercial", "SmolLM-135M / sub-70ms / 0 output tokens", "demo P(True) 0.5052 / Choice conf 0.2872 / Score conf 0.0055", "README claims MIT / GitHub license null / no LICENSE file", "patelvishwa112/jev-system-one-rlcd ≠ arnabgho/rlcd-lite ≠ blackwood-rlcd", "source-backed Awesome Jev radar / 306+ commit-pinned", "logicrw/awesome-jev-projects ≠ AnotiaWang/awesome-jev ≠ yibie/awesome-jev ≠ cobanov/awesome-jev ≠ rupeshpoojary9/awesome-open-system-one", "auto GitHub sync / Issue-only submissions", "hashed n-gram encoder / rival-aware attention", "olanotolu/jevbetter vs jevlike starter", "synthetic hard menus top-1 0.916 vs 0.873 / ECE 0.0182 vs 0.0367 / 40 vs 4608 menus/sec", "shuffled-context control 0.335", "Turn any open LLM into System-One Jev", "uspraveen/Jevify ≠ Mintzs/jevify ≠ gulagala001/jevify", "Jevify-any-LLM architecture probe", "description-only stub / size 0", "Train encoder-only calibrated decision models from a task sentence", "Exu is a toolkit, not a method", "strictly proper scoring rule", "Pre-alpha", "Ruivalim/exu-base", "scratch-trained calibrated decision model", "typed Q → probability dists", "Colvin0315/MiniSystemOne ≠ hyusi2003/MiniSystemOne", "no published weights download URL", "90.5 seconds / 29.2% pipeline evidence", "p_i/p_j independent of other candidates", "Recipe for calibrated decision models — small model out", "init → synth → train → eval → serve", "91.1 % / ECE 0.022 *theirs*", "Jev zero-shot 75.1", "scienthoon/luce", "Put Jev's three headline claims on trial", "0.5B local GPU", "46x speedup / accuracy identical", "ECE 0.624 sentiment catastrophe", "bigger model worse calibration", "RichardoMrMu/jev-mini ≠ yuki-oshio/mini-jev ≠ r-ms/mini-jev", "System-1 decision engine for local LLMs", "structured choices only", "JSON parse of generated text ≠ Noul", "TypefAI JEV / Journal Entry Voucher", "tapsin/jev-local ≠ us/jev-local ≠ Argos1111/jev_local", "Jev 1.13 reward-model eval across 8 benchmark tracks", "40,940 examples / 0 API errors", "RewardBench v1 92.58%", "Precise IF 50.63%", "goya4140/jev-reward-model-evaluation", "Scaffolding in progress", "Jev vs LLM support-ticket routing", "static + live decision bench", "TypeSafe's own published benchmark", "illustrative simulations, not live API calls", "JevBench v1 — smart/cheap/fast/reliable", "I/C/S/K 25% geometric mean", "classifier.dev fast tier 84.8 is Jev behind its own API", "do not re-fold §78 v1.2 board as new", "Laya (421M) 70.1 now on board", "Zero-shot/few-shot LLM routing", "hard budget filter before Jev", "Jev never asked to perform budget arithmetic", "Jev judges the next state, XState enforces transitions", "simulation uses synthetic keyword fixtures", "catalog gravity", "v-modal/awesome-jev-tools", "★339 live REST", "curation is not endorsement", "crawler-maintained directory", "Daily GitHub + npm sweep, human-merged", "RadRebelSam/awesome-jev ≠ AnotiaWang ≠ yibie ≠ cobanov ≠ logicrw ≠ v-modal", "HF peft SPLADE/BGE reranker", "rdxtremity/jev-reranking ≠ carlaiau/jev-reranking", "query-side encoders, not a Jev replica", "ONNX System One Qwen3.5-4B scorer", "source:pngwn/system-one-qwen3.5-4b-scorer", "CC-BY-NC-4.0", "temperature 1.75", "transformers.js AutoModel cannot load this graph", "Consistency benchmark Space", "This Space contains no benchmark result yet", "12-case plumbing fixture", "Benchmark-driven Jev router and judge", "cheap alone is not success", "Jev does not write, sum prices, or claim accuracy %", "Sol 94.2 / Luna 83.9 / Jev path 89.7", "19.2% Sol / 62.3% cost save / 4.5pp miss of 2pp non-inferiority", "p50 latency worse than Sol due to routing overhead", "erendikmenn/jev-llm-router-benchmark ≠ jev-rag-benchmark ≠ ryantsai/jev-llm-router", "Express + node:sqlite", "mock and Jev decision engines", "previous_ticket_count >= 3 is code", "MIN_CONFIDENCE 0.6 still soft", "substring false positives", "aesaganda/jev-ticket-router ≠ SarathChandraBellam/jev-vs-llm-ticket-router", "Universal Figure & Diagram Router", "confidence ≥ 0.85 hard-gate is theater", "generative AI banned from scientific plots", "six visual branches", "hoangngochuong24947-gif/jev-figure-router", "human-labeled (state, question, label)", "166,054 rows / 22 configs", "soft_label for human uncertainty", "Praveenrajus/jev-bench ≠ fstandhartinger/jevbench", "ternary bonsai System One GGUF", "openjev's mechanism, Bonsai's weights", "Hub does not ship weights", "100/100 easy T/F is not Harbor", "label_mass ≠ correctness", "stock llama.cpp Q2_0 silently gibberish", "NicolaiMTLassen/open-bonzi-jev ≠ NicolaiLassen", "transformers.js DeBERTa ONNX", "source:com-kotobalabs/open-jev-deberta-v3-large", "temperature 1.05", "AutoModel from_pretrained works", "onnx-community/open-jev-deberta-v3-large-ONNX ≠ system-one-qwen3.5-4b-scorer-ONNX", "107★ densify", "GH 151M vs README 149.6M", "PR #1 now closed unmerged", "do not re-fold §71 claim-audit as a beat", "typed decisions, RLCD, confidence-gated routing", "structured ≠ correct", "mock not live API", "26 tests", "wjdjdakf17/jev-study ≠ baekenough/jev-study", "bonzi-27b-v2 / ternary-8b / 27b-v1 GGUF family densify", "WANLI-256 74.6% / 65.2% / 71.1% *theirs*", "Bonsai 1 27B Q1_0 runs on stock llama.cpp", "ternary still needs PrismML fork", "hf:heman10x/openJev-verdict-2.0 twin tokenizer-only", "OpenJev Vision image classification + uncertainty", "CLEVR-4 held-out joint 0%", "hfdataset:IamBusy/OpenJev-Vision-Research-v0.1 12,832", "294,912 derived targets not independent samples", "Laya multilingual ONNX WebGPU typed-decisions port", "63/63 selected answers / 5.1e-4 CPU / 1.2e-2 WebGPU", "UpHash-Network/mini-jev is yuki-oshio transfer", "jev-injection-bench 11,900 labelled prompts", "Jev best ranking / Haiku better ECE 0.021 vs 0.058", "0.5–0.9 band is where Jev's numbers do not mean what they say", "Prompt wording moves panic 28%", "manojlds/jev-dspy-bench ≠ dspachos/jev-dspy ≠ jmanhype/jev-dspy-lab", "Jev agreement is similarity, never ground truth", "no aggregate quality grade or merge gate", "AbstentionBench-on-Jev rank 1 of 20 vs 2025 field", "question-asymmetry", "forward-looking 0.465 never extreme", "openkev calibration layer not a runtime", "ECE vs coverage independent", "select_threshold returns inf", "escalation catches uncertainty not ignorance", "misakaikato/openkev ≠ jaredpalmer/kev", "pdf-race Docling→Jev vs Gemini", "parser owns the wall clock", "12/12 tie is a tie", "titles selected not generated", "flopcheck 16 calibrated tweet judgments", "mechanical tells in code", "ZeroX-01/jev-atlas ≠ Zaious/jev-capability-atlas ≠ gorock007/jev-atlas", "Laya calibration lab Gradio MCP", "T never changes argmax", "confidence ≠ top-label p", "easy probe set refused", "40–48 rows too small to ship T", "Gemma-4 26B-A4B jevify classification+calibration", "LoRA adapter twin not independent eval", "Gemma-4 E4B jevify", "E4B LoRA stub card", "kushalpatil/jevify-gemma4 ≠ Mintzs/jevify ≠ gulagala001/jevify ≠ uspraveen/Jevify", "GH kushalpatil07/jevify 404", "PAWS 0.580/ece 0.288 is the weak cell", "smaller E4B slightly better OOD ECE than 26B-A4B", "Hub jevify merged LoRA ships weights", "bonzi Bonsai-8B v1 GGUF densify", "Bonsai-1.7B v1", "Bonsai-4B v1", "WANLI-256 64.5% / 60.2% / 52.0% *theirs*", "rank #4 / #5 / #6 of 6", "JulesHuisman/jev-eval scaffolding / README SHA c356a584 (was empty e69de29b)", "JulesHuisman/jev-eval ≠ SivletLabs/jev-eval ≠ willkelly/jev-evaluation ≠ 4esv ≠ xxkuboxx ≠ onlyoneaman ≠ dayhaysoos/jevals", "7 bands 6/10 vs 40 bands 0/10", "source receipts + confidence slider re-policy without re-inference", "32/32 synthetic is smoke not production", "classify HF datasets across typed semantic dimensions", "roadus2 watch misspelling; lock roadius2/ultra_laya", "ultra_laya REVIEW defects", "default branch claude/laya-jev-review-gg5ppo", "XNLI EN 88.3% ECE 0.032 → RU 77.3% ECE 0.096", "Δ −11.0 pp [−14.2,−7.8]; ECE +0.063", "MASSIVE no detectable difference at n=600", "confidence is function of p_max (r=1.000)", "pointer-not-generator 400 human-authored responses", "proposed ≠ authorized", "FewRel 160: Jev 85.0% vs lexical 13.125%", "gated 100% (95/95) coverage 59.375%", "J++ composable semantic computation language", "judge-jev 0.5 still soft", "947 repos scored; A 273 / B 302 / C 372", "LLM rubric ≠ benches", "No benchmark winner is claimed", "phishing: naive 62.6% vs regex 91.8%; 5-atomic + LR 95.0% *theirs*", "AITuber tension ±15", "README npm global; repo is Rust", "git-confess code owns counting/blame/ratio", "httpx exhibit 11% (13/119) *theirs*", "90d trend +12.40% vs random +12.75% vs BH +41.71%", "5m win rate 25%", "Awesomejev 656 entries / 38,160 stars", "tracker likes 64 (+4) lastModified UNCHANGED", "Laya present; Blackwood ABSENT; Archer still promised_not_landed", "Blackwood tracker ABSENT; likes 2 gated manual", "r = c - p_a", "ECE 0.021; acc 0.807 vs warmup 0.746", "Independent primitive", "11.57s vs 54.10s · 4.67× · 120/128 *theirs*", "default path is pretrained Gemma probs not trained RLCD head", "GH Meanblock 404; lock leesk212/JEV-CPU", "softmax over letter slots ≠ Noul", "WANLI 0.741 vs openjev v2 0.77 *theirs*", "3-way NLI ≠ Noul", "priority 0.464 = majority floor", "banking77 contaminated", "raw margins not probabilities", "do not distill Jev as teacher of record (they distilled Haiku)", "“0.9 is not one number”", "ranking ≠ calibration", "banking77 0.8–0.9 stated 0.86 actual 0.73 over-confident *theirs*", "≠ Praveenrajus/jev-bench ≠ fstandhartinger/jevbench", "$0.0000153–$0.0000226 vs circulating $0.0004 (~20×)", "Score is 0..n-1 expectation not 0–1", "Noul has no confidence field", "TCP floor 198.8 ms", "type reliability is not a reason to choose Jev (json_schema 5/5)", "gateway tax not one number", "Function-only 5/8 vs hybrid 8/8", "4/8 without Jev", "8 designed cases not conversion lift", "200-row pilot Jev 86.5% 173/200 vs Gemini Flash-Lite 86.0% 172/200 vs Pro 87.0% 174/200 *theirs*", "not a ranking", "情緒測謊器", "8-example Jev vs GPT-5.6 Sol ~64× cost 5.4× latency *theirs*", "synthetic; no inference", "≠ JevBench v1.2 §78", "Judged 3317 / listed 2560", "Jev judges, code applies policy", "APA “microsecond policy / zero hallucination” overclaim", "Client-side quiz; pointer from held docs; scanned-PDF warn", "Jev judges / agent reasons / user decides", "selecting an option is not permission to implement", "pattern exact, judgement must clear floor", "no matching pattern → no model call", "not a correctness oracle", "Spec vs artifact remainder", "treating 0.85 as 85% / minProbability hard-gate as Harbor", "VERIFY acquires discriminating evidence, never same-pool confidence-only rescoring", "fast/full/max are ceilings not sizes", "Solar writes, Jev chooses NEXT ACTION", "do not reopen or amend PR #23 or #24 or #25 or #26 or #27", , "Calibration is not alpha", "NO CURRENT ALPHA CANDIDATE", "ΔR² approximately +0.00084", "Brier 0.2131387", "ECE 0.0421875", "Adding Jev probability to deterministic volatility improved Brier by only 1.4058e-05", "default 0.5 keeps zero non pinned", "keepResult median 0.14 to 0.17", "keepCall median 0.28 to 0.35", "usable range is about 0.10 to 0.25", "7.8% to 57.9%", "judges results it never sees", "task-finish eval not built yet", "$0.002 per compaction", "slavadubrov/sgr-judge-bench ≠ slavadubrov/jev-judge-bench", "Jev 108/120 $0.083 0.34 s", "Luna SGR 114/120", "paired Jev accuracy-difference intervals include zero", "not evidence of equivalence", "GLM SGR 26/120 93 format failures", "Terra-planned Jev hybrid 55/120", "rule-based by default, optionally Jev-backed", "empty README", "missing key cannot break the experience", "prefill plus exactly one decode", "softmax over A/B/C ≠ Noul", "BBQ 9,053/10,000 (90.53%)", "ECE 0.0890", "Mean confidence 0.9943", "overconfident", "score and noul not implemented", "DGUI 12 rows (was 6)", "INSTRUCT 119 rows likes 2", "encode the state once, decide everything in parallel", "0.740 accuracy against a 0.508 majority", "ECE 0.047", "fine-tune's advantage ends where its 384-token training data does", "jasonkneen/open-jev ≠ pngwn/open-jev", "same sha d41dc3cd", "Space does not call Jev", "recomputes routing from saved probabilities", "200-case Jev 97.0% / 100.0% / 95.0% / MAE 9.22", "synthetic repository benchmark", "Jev evaluations are advisory", "YehuiTang0316/jev-nlgrep ≠ Bentlybro/jevgrep ≠ can1357/jegrep ≠ uehaj/jev-semgrep", "default threshold 0.8 still soft", "40-line windows cannot prove whole function", "token-native sequential start/end Choice", "Gemini/Haiku stubs not configured yet", "handful of hand-written examples, not a benchmark", "Jev judged exactly what it was given", "laguagu/jev-skills ≠ laguagu/jev-evidence-lab ≠ Pleo2/awesome-jev-agent-skills", "contract_passed is not a claim of guaranteed factual truth", "Wilson lower bound 0.85 floor", "fixture mode no savings claim", "SemIf 2207★ (+21 vs §110 2186)", "jevlike 1043★ (+5 vs 1038)", "TypeAR 15★ (+1 vs 14)", "AnotiaWang 97★ (+1 vs 96)", "yibie/awesome-jev 506★ (+16 vs 490)", "Laya likes 822 (was 802)", "tracker likes 64 flat, lastModified UNCHANGED", "do not reopen or amend PR #23/#24/#25/#26/#27/#28", "Heman10x-NGU/Verdict-open-jev ≠ Heman10x-NGU/openJev-verdict-2.0", "TF-IDF + LogReg ECE 0.0207 vs Jev 0.1440", "Verdict-open-jev 48.07% vs Jev 90.80%", "abstention combined recall 10.00%", "p50 35.58 ms", "K=25 (maximum capacity) 72.00%", "0.85 coverage 84.60% selective risk 1.18%", "26.1× faster than standard Qwen JSON generation", "Jevify 90.0% / 167 ms CUDA graphs disabled", "Finding 1: Brier on stated confidence alone is a trap", "grpo_rlcr 0.78 / ECE 0.084", "reliability 0.007 but resolution 0.000", "27 900 schema-driven decisions", "13 600 / 13 600 questions", "candidate mass min 0.99999624", "22 configs · 166,054 rows · 4 calibration-gold", "sha a39eba3f", "Student B MAE 0.148 / Pearson 0.836 / 86.0%", "pngwn/open-jev-laya-bench README 404", "sha 9f69c742 likes 2", "HDFS 0.9933 (745/750) / retain 0.0084", "BGL ERROR/FATAL protection 1.0000", "2,479 / 2,500 HDFS uncertain", "cache hit 0.9648 (2412/2500)", "$0.153936 estimated", "E2 recomputes from saved probabilities", "Space sha eda59e0a", "MASSIVE English 0.783 / Khmer 0.033 / Hindi 0.133", "40–48 rows too small to ship T", "T never changes argmax", "siren2345/jev-single-decode-transformers ≠ siren2345/jev-single-decode", "Split Transformers experiment from llama.cpp runtime", "tanayvasishtha/jev-lab ≠ dairui1/jev-lab ≠ BrendanH18/jev-lab ≠ yibie/laya-jev-lab", "Four experiments stress-testing TypeSafe's Jev: calibration, bundle bias, label bias, and ensembling", "second pass must be $0.00 from cache", "The pages never call Jev", "Gemma 4 31B 77.0% / Jev 1.13.0 61.4% / Laya 322M 0.0%", "restriction state 95.0% against 84.4%", "None of the systems are particularly good at knowing when to stop and ask", "They skip the question and call a tool directly", "100% schema pass", "six-field joint 48.8% vs 72.8%", "ywchiu/jev_benchmark ≠ Running-Dolphins/jev-bench ≠ Praveenrajus/jev-bench", "ACT / REVIEW / FALLBACK", "A provider failure, timeout, malformed output, or missing answer is **not** a policy outcome", "confidence is descriptive provider output, not a substitute for probability", "Quality denominators include only valid scored answers", "an exact halfway tie chooses the lower level", "aiwithenoch/Jev-Skill ≠ simplosophy/jev-skill ≠ laguagu/jev-skills", "The local path does not claim to turn a smaller checkpoint into Jev", "Low support becomes decision: \"review\"", "MIT-0 SPDX NOASSERTION", "current-llm", "结构兼容，不是 Jev 模型能力", "altryne/jevify ≠ Mintzs/jevify ≠ gulagala001/jevify ≠ uspraveen/Jevify", "Find where Jev belongs. Design the questions. Measure the difference", "TypeAR-AI/TypeAR 301 → TypeLLM/TypeLLM", "TypeLLM/TypeLLM 16★", "SemIf 2241★ (+34 vs §111 2207)", "jevlike 1051★ (+8 vs 1043)", "AnotiaWang 98★ (+1 vs 97)", "yibie/awesome-jev 525★ (+19 vs 506)", "Laya likes 864 (was 822)", "tracker likes 67 (+3 vs 64)", "lastModified UNCHANGED `2026-09-20T04:29:16.000Z`", "do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#32", "hysteresis enter/exit / replay policy without inference", "calibration does not compose / hop-ECE permutation-invariant", "equal-width vs quantile ECE / ranking ≠ calibration", "Qwen2.5 ≠ Archer / Qwen 3.8 sparring ≠ Archer / Qwen/Qwen3.8-27B ≠ Archer", "Deferred Crispification / TCE / AMS", "g0runmezadam/what-is-jev IS tunahansahin897/what-is-jev", "pd.cut equal-width vs jeval quantile", "A hunch is a probability with a policy attached", "soundness theater / measurement theater / hourly 0843", , "Jev Capability Resolver / NiazMorshed2007/jcr", "one tool nested capability tree / returns context / does not execute", "skills vs capabilities / workflow+judgment vs operations", "format independent of Jev / proposed open standard", "JCR_BAND_RATIO 0.6 is application policy / soft scores ≠ hard gates", "routing ≠ permission / docs ≠ authority to run", "sol-vs-opus5-20 lookup+explain / n=1 / Not Harbor task-execution", "wall-time mixed / Sol slower with JCR in 19/20", "NiazMorshed2007/jcr ≠ skill-broker ≠ skillranker ≠ jev-sift ≠ jev-lens ≠ jevusher ≠ jev_select_capability", "do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34", "notes.md §116", "copy the SemIf/MLX installer?", "quote 5.21× as beating Jev?", "treat 0.845 as a TypeSafe replica?", "collapse SemIf into kw2828/zhihz/semif-rs/semif-serve", "softmax over options as a Noul", "llm prompt to jev primitives", "conversion assistant not equivalent behavior", "heuristic conversion ≠ calibrated Noul", "alexwestco/llm-to-jev ≠ altryne/jevify", "user-provided 0940 / notes.md §118", "judge ≠ actuator", "candidate_mass", "softmax over A–H ≠ Noul", "hourly 0947 / notes.md §119", "ggmlc GGUF is not llama.cpp", "serving substrate ≠ calibrated replica", "Qwen3.5-9B ≠ Archer", "planner writes JEV selects", "hourly 1049 / notes.md §120", "open recreation ≠ calibrated replica", "semantic lint is a sensor not a proof", "cutoff 0.8 still soft", "paired bootstrap CIs *theirs*", "Same accuracy, 35x faster *theirs*", "hourly 1143 / notes.md §121", "revisit HIGH / since-last-look", "catalogued repo changed", "star-noise vs material change", "densify prior notes without inventing equivalence", "decide is not generate", "tryDecide returns typed calibrated judgments not a token stream", "GLiNER/GLiClass ports are class members not Jev replicas", "93.5% *theirs* not Harbor", "74.9 *theirs* not Harbor", "8.7x *theirs* not Harbor", "Option-Marker joint attention", "openjev:0.2.1", "thinking=True/False per-field budget", "PLAN_Qwen35", "hyperspaceai/jevcache ≠ kushals256/jevcache", "wire-compat ≠ logit-equiv", "SHA move is not a replica", "hourly 1248 / notes.md §123", "typesafe-sdk 0.7 Pydantic response models", "msgspec dropped", "The server's output is unchanged and was never wrong", "SchemaError is 400 plain-string detail not 422 list", "Pydantic response models ≠ logit-equiv", "msgspec dropped is not a replica", "Error contract is not a Noul", "coverage-at-error-budget *theirs* not Harbor", "PLAN_Qwen35 still proposal for review", "GLiNER locate ports are class members not Jev replicas", "Locate ≠ decide", "~160 ms *theirs* not Harbor", "0.971 F1 *theirs* not Harbor", "hf:fr0stbit3/laya-gguf serving substrate ≠ calibrated replica", "jkcdarunday/SystemOne-Next ≠ TypeSafe System One", "hourly 1340 / notes.md §124", "vLLM NVIDIA + MLX Apple Silicon", "Codiv hosted free endpoint", "dual /v1/systemone + /v1/chat/completions", "chat 501 on MLX", "dual serving is not generate", "Hosted Codiv ≠ TypeSafe", "hr98w/jev-visual 167★ Apple Silicon visual candidate scoring", "37.30s → 2.40s at 64 decisions *theirs*", "Breakout 9 bricks 6 returns 2 lives *theirs*", "candidate probabilities are relative not correctness", "jkudish/jev-mcp 156★ ten MCP tools", "recommendation is advisory", "the server never blocks on its own", "TypeSafe CLERC 5% to 18% *theirs*", "jkudish/jev-mcp ≠ burnigtm/jev-mcp", "zhengxuyu/litjev off-the-shelf Qwen decision layer", "Probabilities are not calibrated by default", "Qwen/Qwen3.8-27B ≠ Archer", "zhengxuyu/litjev ≠ alexwestco/llm-to-jev", "Zefan-Cai/Open-Jev LoRA + scalar head", "2B 94.71% 9B 97.54% hard test *theirs*", "2B OOD 86.02% 9B OOD 91.97% *theirs*", "80,816 training rows", "27B still in progress", "LoRA ≠ RLCD replica", "Zefan-Cai/Open-Jev ≠ TheoLeeCJ/openjev ≠ razorback16/openjev", "cristianoliveira/jeq intelligence you can pipe", "pass-min 0.8 still soft", "JEQ does not own actions", "AndyInQtr/laya-coreai CoreML serving substrate ≠ calibrated replica", "AndyInQtr/laya-coreai ≠ mizorewww/laya-coreml", "hourly 1441 / notes.md §125", "TypeLLM/TypeLLM densify HEAD 6a48f9f1e623", "README densify 3k→12k B", "Batch 5.8x *theirs*", "Constrained AR ≠ calibrated Noul", "jaredpalmer/kev densify HEAD b339f446a0ef", "Kev-0.6B 4B 8B family", "4B new-source 0.790/0.806 *theirs*", "8B new-source 0.796/0.780 *theirs*", "Jev hosted 0.857 *theirs*", "Questions share the input text but cannot read each other", "No Jev outputs were used for training", "8.2% ≥0.9 on wrong *theirs*", "option order can change an answer", "Qwen3 ≠ Archer", "TheoOliveira/pi-jev 21★ fail-closed routing", "JEV_THRESHOLD 0.65 still soft", "harshwasan/jev-sentinel fail closed never auto-allows", "harshwasan/jev-sentinel ≠ leepokai/jev-guard", "jackbarunz/jev-tool-router ≠ esinocchi/jev-tool-router", "threshold 0.90 still soft", "76/81 vs 77/81 *theirs*", "0.419s vs 2.459s *theirs*", "$0.00486 vs $0.03673 *theirs*", "not a security boundary", "baronunread/leanest fail-open uncertainty means RUN", "classifier.dev default Jev/Laya pluggable", "openlayer-ai/jevals ≠ dayhaysoos/jevals", "estimates not Harbor", "classifier ≠ authorizer", "MrJev/awesome-jev 118 entries catalog ≠ endorsement", "MrJev/awesome-jev ≠ yibie/awesome-jev", "Koushik890/jev-firewall fail closed ask_below 0.7 still soft", "CompleteTech-LLC-AI-Research/jev-codex-approval experimental native not compiled", "confidence is not a measured probability", "rh-guard owns primary gates", "hf:rAVEUK/open-jev-deberta-v3-large encoder class member not Jev replica", "hf:p-yan/laya-quanto serving substrate ≠ calibrated replica", "hf:Gtrkrsk/laya serving substrate ≠ calibrated replica", "hourly 1542 / notes.md §126", "razorback16/openjev densify HEAD febf02e88989", "release 0.3.0", "re-pin vLLM PR #57250 restructured head", "MODEL_VERSION stays openjev-0.1", "uv.lock hygiene", "restructured vLLM head ≠ logit-equiv", "frostney/clean-code-review 7★ typed judgments not opinions", "documentation is read not judged", "morcoan/JMP Joint Model Participation", "Models participate. Real tools execute.", "Thresholds are policy not model", "Kelbie/hunch ≠ carldaws/hunch ≠ tpellet/hunch ≠ huncho", "Jev never generates prose JSX or code", "json-render is the only renderer", "game success ≠ calibrated Noul", "Shalimov04/open-jev ≠ razorback16/openjev", "MstyAI/laya-onnx empty repo", "hf:Praveenrajus/jev-bench HTTP 200 was 401", "hourly 1643 / notes.md §127", "TypeLLM/TypeLLM densify HEAD 702e6a287f3c", "truncated thinking then constrained decode", "0.8B thinking On 0/18 *theirs*", "forced closure 20/20 type-valid *theirs*", "jaredpalmer/kev densify live HEAD 8465c4c4c294", "Kev-0.8B completes family", "4B new-source 0.794/0.832 *theirs*", "9B new-source 0.812/0.837 *theirs*", "transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*", "SemIf Kev-9B 0.917 Jev 0.965 *theirs*", "scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*", "transformers >= 5.17", "Qwen3.5 ≠ Archer", "notque/vexjoy-agent 421★ /d routes /do fallback", "Facts go to code. Judgments go to Jev. Only facts can block.", "Jev never blocks", "jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev", "five-lines threshold 0.80 still soft", "371ms $0.0000189 300-call *theirs*", "tpellet/jevify ≠ altryne/jevify", "seb4ez/jevguard-mcp ≠ seb4ez/jevguard", "resumocast/jev-mcp ≠ jkudish/jev-mcp", "Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort", "MidasMulli/kev-ane 155/155 argmax *theirs*", "hourly 1746 / notes.md §128", "Fine-tuning on your own data", "--data JSONL", "--init_from warm-start LoRA/head PR #9", "from-scratch ≠ warm-start", "JSONL labels ≠ Harbor", "Kev-0.8B 4B 9B Qwen3.5 family", "0.33 vs 0.84 vs 0.83/0.88 *theirs*", "reconstruction ≠ replica", "assay-001 split verdict", "Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev", "ThePFMind/jev-mcp ≠ jkudish/jev-mcp", "kyegomez/open-jev ≠ razorback16/openjev", "namenu/pi-jev-effort ≠ TheoOliveira/pi-jev", "samatv256/mini-Jev ≠ r-ms/mini-jev", "hourly 1843 / notes.md §129", or "cascade sign-flip / calibration theater": read `references/faq.md`, then `references/mental-models.md`, then `references/mixed-architecture.md`, then `references/judgment-class.md` before any mapping. Proof, @@ -543,3 +543,6 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 + +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/agent-self-assessment.md b/.agents/skills/augustus/references/agent-self-assessment.md index fbfe4f1..6f85ffb 100644 --- a/.agents/skills/augustus/references/agent-self-assessment.md +++ b/.agents/skills/augustus/references/agent-self-assessment.md @@ -967,3 +967,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/applied-mappings.md b/.agents/skills/augustus/references/applied-mappings.md index 5ef3375..f864066 100644 --- a/.agents/skills/augustus/references/applied-mappings.md +++ b/.agents/skills/augustus/references/applied-mappings.md @@ -2546,3 +2546,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/composition-algebra.md b/.agents/skills/augustus/references/composition-algebra.md index d7e0731..3d50144 100644 --- a/.agents/skills/augustus/references/composition-algebra.md +++ b/.agents/skills/augustus/references/composition-algebra.md @@ -2978,3 +2978,76 @@ truncated thinking then constrained decode; Constrained AR ≠ calibrated Noul; Facts go to code. Judgments go to Jev. Only facts can block; Jev never blocks; Kev-0.8B completes family; 155/155 argmax *theirs*. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 + +481. **kev own-data JSONL densify PRIMARY** (jaredpalmer/kev): + densify §45. HEAD bd058057ad0a README SHA 84b872488915. 1033★. + Fine-tuning on your own data. --data JSONL. + --init_from warm-start LoRA/head PR #9. + Kev-0.8B 4B 9B Qwen3.5 family. SHA move is not a replica. + Full cards: `judgment-class.md`, `validation.md`. +482. **from-scratch ≠ warm-start / JSONL labels ≠ Harbor** (jaredpalmer/kev): + 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. + 0.33 vs 0.84 vs 0.83/0.88 *theirs*. from-scratch ≠ warm-start. + JSONL labels ≠ Harbor. Kev-0.5B card Qwen3.5 family pointer. + Full cards: `faq.md`, `mixed-architecture.md`. +483. **dabit3 densify already catalogued** (dabit3/jev-experiments): + dabit3/jev-experiments densify 340★. densify is not a sibling first sighting. + Full cards: `applied-mappings.md`. +484. **tenbin densify neighbor skill** (simota/tenbin): + simota/tenbin densify neighbor skill. Augustus does not absorb it. + Full cards: `faq.md`. +485. **reconstruction ≠ replica** (kyegomez/open-jev): + unofficial research implementation with random weights. + reconstruction ≠ replica. + kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev. + Full cards: `judgment-class.md`, `faq.md`. +486. **assay-001 split verdict** (jourdanlabs/assay-001): + assay-001 split verdict. CLINC150 ECE 0.0204 *theirs*. + Banking77 ECE 0.0936 *theirs*. 8,576 responses zero type errors *theirs*. + *theirs* not Harbor. Full cards: `validation.md`. +487. **jev-ra latency *theirs*** (brnyxx/jev-ra): + brnyxx/jev-ra 3-5x / ~300 ms *theirs*. 8.50× Wikipedia *theirs*. + systems comparison ≠ semantic equivalence. + Full cards: `validation.md`. +488. **awesome-jev catalog namesake** (Promethe-us/awesome-jev): + Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev. + catalog ≠ endorsement. Full cards: `faq.md`. +489. **jev-mcp namesake** (ThePFMind/jev-mcp): + ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp. + serving substrate ≠ calibrated replica. Full cards: `faq.md`. +490. **minesweeper cousins** (comoc/jev-minesweeper, EnesYilmazcode/JevMinesweeper): + comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper. + game success ≠ calibrated Noul. Full cards: `validation.md`. +491. **pi-jev-effort namesake** (namenu/pi-jev-effort): + namenu/pi-jev-effort ≠ TheoOliveira/pi-jev. + rh-guard owns primary gates. Full cards: `faq.md`. +492. **HF first-sighting / densify** (mini-Jev / allmix-r512 / laya-*): + samatv256/mini-Jev ≠ r-ms/mini-jev. + hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo. + serving substrate ≠ calibrated replica. + Full cards: `judgment-class.md`, `faq.md`. +493. **remainder apps** (fraud / notion / quilt / wrapper / DriftLab / jev-hft / others): + Nutlope/jev-fraud Kimi K3. jeffloo886/jev-notion. + classifier ≠ authorizer. catalog ≠ endorsement. + Full cards: `applied-mappings.md`, `faq.md`. +494. **namesake remainder** (open-jev / awesome-jev / jev-mcp / mini-Jev / minesweeper): + reconstruction ≠ replica. catalog ≠ endorsement. + Full cards: `faq.md`. +495. **skip-thin catalogs** (jevguide / awesome-jev-prompt / hellojev): + catalog ≠ endorsement. skip-thin. Full cards: `faq.md`. +496. **skip Archer** (promised_not_landed): + Qwen3.5 ≠ Archer. Hub archerhume/4rcherhume HTTP 401. + Archer still promised_not_landed. Full cards: `faq.md`. + +Hourly 1843 items 481–496 (`notes.md` §129). Do **not** +re-fold §128 items 465–480 / §127 items 449–464 / §126 items 433–448. +Skip Archer rewrite. +from-scratch ≠ warm-start; JSONL labels ≠ Harbor; +reconstruction ≠ replica; assay-001 split verdict; +catalog ≠ endorsement; game success ≠ calibrated Noul; +SHA move is not a replica. +do not reopen or amend PR #23–#51. +Soft Noul ≠ hard safety. + +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/faq.md b/.agents/skills/augustus/references/faq.md index 3f4cde4..3083361 100644 --- a/.agents/skills/augustus/references/faq.md +++ b/.agents/skills/augustus/references/faq.md @@ -3752,3 +3752,14 @@ Do not reopen or amend PR #23–#50. `invented_signal: false`. `notes.md` §128. **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 + +## Treat kev --init_from as from-scratch? Treat JSONL as Harbor? Treat kyegomez as a replica? + +No. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. reconstruction ≠ replica. +unofficial research implementation with random weights. assay-001 split verdict. +catalog ≠ endorsement. game success ≠ calibrated Noul. +Do not reopen or amend PR #23–#51. +`invented_signal: false`. `notes.md` §129. + +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/formal-methods.md b/.agents/skills/augustus/references/formal-methods.md index 95764a9..829ec4e 100644 --- a/.agents/skills/augustus/references/formal-methods.md +++ b/.agents/skills/augustus/references/formal-methods.md @@ -1396,3 +1396,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/formal-semi-formal.md b/.agents/skills/augustus/references/formal-semi-formal.md index 00362e1..9c4f0da 100644 --- a/.agents/skills/augustus/references/formal-semi-formal.md +++ b/.agents/skills/augustus/references/formal-semi-formal.md @@ -109,3 +109,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/judgment-class.md b/.agents/skills/augustus/references/judgment-class.md index 006776a..8630e68 100644 --- a/.agents/skills/augustus/references/judgment-class.md +++ b/.agents/skills/augustus/references/judgment-class.md @@ -1510,3 +1510,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/mappings.md b/.agents/skills/augustus/references/mappings.md index edf75b7..386dbbc 100644 --- a/.agents/skills/augustus/references/mappings.md +++ b/.agents/skills/augustus/references/mappings.md @@ -2501,3 +2501,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/mental-models.md b/.agents/skills/augustus/references/mental-models.md index ce2da35..be3ef3c 100644 --- a/.agents/skills/augustus/references/mental-models.md +++ b/.agents/skills/augustus/references/mental-models.md @@ -3034,6 +3034,17 @@ Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. serving substrate ≠ calibrated replica. SHA move is not a replica. Do not copy keys. +## Apply 1843 (`notes.md` §129) + +from-scratch ≠ warm-start. JSONL labels ≠ Harbor. +--init_from warm-start LoRA/head PR #9. Fine-tuning on your own data. +Kev-0.8B 4B 9B Qwen3.5 family. 4B new-source 0.794/0.832 *theirs*. +9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. +reconstruction ≠ replica. unofficial research implementation with random weights. +assay-001 split verdict. CLINC150 ECE 0.0204 *theirs*. +Banking77 ECE 0.0936 *theirs*. catalog ≠ endorsement. +game success ≠ calibrated Noul. SHA move is not a replica. Do not copy keys. + ## Apply 1643 (`notes.md` §127) restructured vLLM head ≠ logit-equiv. release 0.3.0. @@ -3100,3 +3111,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/methods-catalog.md b/.agents/skills/augustus/references/methods-catalog.md index 5239502..3c46c44 100644 --- a/.agents/skills/augustus/references/methods-catalog.md +++ b/.agents/skills/augustus/references/methods-catalog.md @@ -303,3 +303,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/mixed-architecture.md b/.agents/skills/augustus/references/mixed-architecture.md index 8d5f279..84219fb 100644 --- a/.agents/skills/augustus/references/mixed-architecture.md +++ b/.agents/skills/augustus/references/mixed-architecture.md @@ -1481,3 +1481,5 @@ Apply 1746 (`notes.md` §128): Facts go to code. Judgments go to Jev. Only facts **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/question-design.md b/.agents/skills/augustus/references/question-design.md index 5617e1f..9463907 100644 --- a/.agents/skills/augustus/references/question-design.md +++ b/.agents/skills/augustus/references/question-design.md @@ -450,3 +450,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/toolbox-mapping.md b/.agents/skills/augustus/references/toolbox-mapping.md index 0a19edf..3097335 100644 --- a/.agents/skills/augustus/references/toolbox-mapping.md +++ b/.agents/skills/augustus/references/toolbox-mapping.md @@ -379,3 +379,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/references/validation.md b/.agents/skills/augustus/references/validation.md index d7dc67c..bebaa3c 100644 --- a/.agents/skills/augustus/references/validation.md +++ b/.agents/skills/augustus/references/validation.md @@ -1185,3 +1185,5 @@ Apply 1746 (`notes.md` §128): 0.8B thinking On 0/18 *theirs*; forced closure 20 **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/.agents/skills/augustus/scripts/evaluate_decisions.py b/.agents/skills/augustus/scripts/evaluate_decisions.py index 9b441ad..3f3a27c 100755 --- a/.agents/skills/augustus/scripts/evaluate_decisions.py +++ b/.agents/skills/augustus/scripts/evaluate_decisions.py @@ -466,6 +466,27 @@ def five_lines_threshold_still_soft(threshold, hard_gate=False): raise ValueError("unexpected threshold") return hard_gate is False +def from_scratch_is_not_warm_start(init_from, from_scratch=False): + """--init_from warm-start LoRA/head is not from-scratch.""" + if init_from != "warm_start": + raise ValueError("unexpected init") + return from_scratch is False + + +def jsonl_labels_are_not_harbor(kind, harbor=False): + """JSONL labels ≠ Harbor.""" + if kind != "jsonl_labels": + raise ValueError("unexpected kind") + return harbor is False + + +def reconstruction_is_not_replica(kind, replica_claimed=False): + """from-first-principles reconstruction ≠ replica.""" + if kind != "reconstruction": + raise ValueError("unexpected kind") + return replica_claimed is False + + def hop_ece_permutation_invariant(rows, bins=10, key="p"): """Shuffle order; equal-width ECE must not move. @@ -664,6 +685,16 @@ def self_test(): assert constrained_ar_is_not_noul("constrained_ar", False) + # 1843: own-data JSONL / from-scratch ≠ warm-start / reconstruction ≠ replica. + assert from_scratch_is_not_warm_start("warm_start", False) + assert not from_scratch_is_not_warm_start("warm_start", True) + assert jsonl_labels_are_not_harbor("jsonl_labels", False) + assert not jsonl_labels_are_not_harbor("jsonl_labels", True) + assert reconstruction_is_not_replica("reconstruction", False) + assert not reconstruction_is_not_replica("reconstruction", True) + assert theirs_bench_is_not_harbor(836, "kev-init-from-836") + assert theirs_bench_is_not_harbor(2, "assay-001-split") + print("self-test ok") diff --git a/.agents/skills/augustus/scripts/uniqueness_gate.py b/.agents/skills/augustus/scripts/uniqueness_gate.py index de9d675..78384c7 100644 --- a/.agents/skills/augustus/scripts/uniqueness_gate.py +++ b/.agents/skills/augustus/scripts/uniqueness_gate.py @@ -2,7 +2,7 @@ """Uniqueness gate for merged 0843 (§114), merged 0915 NanoJev (§115), merged 0920 jcr (§116), merged 0922 SemIf (§117), merged 0940 llm-to-jev (§118), hourly 0947 HIGH (§119), hourly 1049 HIGH (§120), -hourly 1143 HIGH (§121), hourly 1248 HIGH (§123), hourly 1340 HIGH (§124), hourly 1441 HIGH (§125), hourly 1542 HIGH (§126), hourly 1643 HIGH (§127), and hourly 1746 HIGH (§128). +hourly 1143 HIGH (§121), hourly 1248 HIGH (§123), hourly 1340 HIGH (§124), hourly 1441 HIGH (§125), hourly 1542 HIGH (§126), hourly 1643 HIGH (§127), hourly 1746 HIGH (§128), and hourly 1843 HIGH (§129). Each lock must appear as one consecutive substring in every listed overlay. Fragments scattered across files do not count. @@ -11,9 +11,11 @@ substring in the skill + research files (not a 21-overlay dump wall). Hourly must treat revisit HIGH like novel HIGH. Star-noise is not a fold. -Also: YAML-parse SKILL.md frontmatter; notes.md owns §114–§128; -composition items 289–316, 322–329, 330–336, 337–352, 353–368, 369–384, 385–400, 401–416, 417–432, 433–448, 449–464, and 465–480 exist; -findings batches #97–#110 exist. Items 317–321 stay unused. +Also: YAML-parse SKILL.md frontmatter; notes.md owns §114–§129; +composition items 289–316, 322–329, 330–336, 337–352, 353–368, 369–384, 385–400, 401–416, 417–432, 433–448, 449–464, 465–480, and 481–496 exist; +findings batches #97–#111 exist. Items 317–321 stay unused. +The 1843 archive run_digest must claim §129 / 481–496 / #111 +(not the 1746 IDs §128 / 465–480 / #110). CHANGELOG.md must not hold uniqueness dump walls (dumps live in changelog-hourly.md). README.md must not hold the 0743 dump wall. Pages greps stay in docs/index.md and docs/_layouts/default.html. @@ -22,6 +24,7 @@ """ from pathlib import Path +import json import subprocess import sys import yaml @@ -158,6 +161,10 @@ 'Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128' ) +UNIQ_1843 = ( + "Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129" +) + REVISIT_LOCK = ( "Revisit / since-last-look lock: catalogued repos are not done; " "store fingerprints default_sha, pushed_at, description_hash, release_tag; " @@ -249,6 +256,8 @@ def main() -> int: failed.append(f"1643 lock missing as one substring: {rel}") if UNIQ_1746 not in body: failed.append(f"1746 lock missing as one substring: {rel}") + if UNIQ_1843 not in body: + failed.append(f"1843 lock missing as one substring: {rel}") for rel in REVISIT_OVERLAYS: path = ROOT / rel if not path.is_file(): @@ -288,10 +297,12 @@ def main() -> int: failed.append("notes.md missing §127 heading") if "## 128. Hourly 1746 HIGH" not in notes: failed.append("notes.md missing §128 heading") + if "## 129. Hourly 1843 HIGH" not in notes: + failed.append("notes.md missing §129 heading") algebra = (ROOT / ".agents/skills/augustus/references/composition-algebra.md").read_text( encoding="utf-8" ) - for n in list(range(289, 317)) + list(range(322, 330)) + list(range(330, 337)) + list(range(337, 353)) + list(range(353, 369)) + list(range(369, 385)) + list(range(385, 401)) + list(range(401, 417)) + list(range(417, 433)) + list(range(433, 449)) + list(range(449, 465)) + list(range(465, 481)): + for n in list(range(289, 317)) + list(range(322, 330)) + list(range(330, 337)) + list(range(337, 353)) + list(range(353, 369)) + list(range(369, 385)) + list(range(385, 401)) + list(range(401, 417)) + list(range(417, 433)) + list(range(433, 449)) + list(range(449, 465)) + list(range(465, 481)) + list(range(481, 497)): needle = f"{n}. **" if needle not in algebra: failed.append(f"composition-algebra missing item {n}") @@ -315,9 +326,31 @@ def main() -> int: "## Batch #108", "## Batch #109", "## Batch #110", + "## Batch #111", ): if batch not in findings: failed.append(f"findings.md missing {batch}") + digest_path = ROOT / "research/archive/hourly/2026-09-21T00/run_digest.json" + if not digest_path.is_file(): + failed.append("missing 1843 run_digest.json") + else: + digest = json.loads(digest_path.read_text(encoding="utf-8")) + if digest.get("label") != "1843": + failed.append(f"1843 run_digest label {digest.get('label')!r} != '1843'") + if digest.get("notes_section") != "129": + failed.append( + f"1843 run_digest notes_section {digest.get('notes_section')!r} != '129'" + ) + if digest.get("composition") != "481-496": + failed.append( + f"1843 run_digest composition {digest.get('composition')!r} != '481-496'" + ) + if digest.get("findings_batch") != 111: + failed.append( + f"1843 run_digest findings_batch {digest.get('findings_batch')!r} != 111" + ) + if digest.get("invented_signal") is not False: + failed.append("1843 run_digest invented_signal is not false") skill = (ROOT / ".agents/skills/augustus/SKILL.md").read_text(encoding="utf-8") try: fm = load_skill_frontmatter(skill) @@ -513,6 +546,33 @@ def main() -> int: 'Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort', 'MidasMulli/kev-ane 155/155 argmax *theirs*', 'hourly 1746 / notes.md §128', + "jaredpalmer/kev densify HEAD bd058057ad0a", + "Fine-tuning on your own data", + "--data JSONL", + "--init_from warm-start LoRA/head PR #9", + "from-scratch ≠ warm-start", + "JSONL labels ≠ Harbor", + "Kev-0.8B 4B 9B Qwen3.5 family", + "0.33 vs 0.84 vs 0.83/0.88 *theirs*", + "Kev-0.5B card Qwen3.5 family pointer", + "reconstruction ≠ replica", + "unofficial research implementation with random weights", + "assay-001 split verdict", + "CLINC150 ECE 0.0204 *theirs*", + "Banking77 ECE 0.0936 *theirs*", + "8,576 responses zero type errors *theirs*", + "brnyxx/jev-ra 3-5x / ~300 ms *theirs*", + "8.50× Wikipedia *theirs*", + "Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev", + "ThePFMind/jev-mcp ≠ jkudish/jev-mcp", + "kyegomez/open-jev ≠ razorback16/openjev", + "namenu/pi-jev-effort ≠ TheoOliveira/pi-jev", + "samatv256/mini-Jev ≠ r-ms/mini-jev", + "comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper", + "hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo", + "dabit3/jev-experiments densify 340★", + "simota/tenbin densify neighbor skill", + "hourly 1843 / notes.md §129", ): if frag not in haystack: failed.append(f"SKILL.md missing fragment {frag!r}") @@ -694,6 +754,21 @@ def main() -> int: 'Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort', 'MidasMulli/kev-ane 155/155 argmax *theirs*', 'hourly 1746 / notes.md §128', + "Fine-tuning on your own data", + "--data JSONL", + "--init_from warm-start LoRA/head PR #9", + "from-scratch ≠ warm-start", + "JSONL labels ≠ Harbor", + "Kev-0.8B 4B 9B Qwen3.5 family", + "0.33 vs 0.84 vs 0.83/0.88 *theirs*", + "reconstruction ≠ replica", + "assay-001 split verdict", + "Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev", + "ThePFMind/jev-mcp ≠ jkudish/jev-mcp", + "kyegomez/open-jev ≠ razorback16/openjev", + "namenu/pi-jev-effort ≠ TheoOliveira/pi-jev", + "samatv256/mini-Jev ≠ r-ms/mini-jev", + "hourly 1843 / notes.md §129", ): if frag not in proto_line: failed.append(f"SKILL.md protocol missing {frag!r}") @@ -713,6 +788,7 @@ def main() -> int: ("1542", UNIQ_1542), ("1643", UNIQ_1643), ("1746", UNIQ_1746), + ("1843", UNIQ_1843), ): if lock in changelog: failed.append( @@ -788,6 +864,7 @@ def main() -> int: f"1542 chars={len(UNIQ_1542)} " f"1643 chars={len(UNIQ_1643)} " f"1746 chars={len(UNIQ_1746)} " + f"1843 chars={len(UNIQ_1843)} " f"revisit chars={len(REVISIT_LOCK)} " f"overlays={len(OVERLAYS)} " f"revisit_overlays={len(REVISIT_OVERLAYS)}" diff --git a/CHANGELOG.md b/CHANGELOG.md index 87efa9f..01768b3 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -16,6 +16,37 @@ folds: `research/notes.md`. ## [Unreleased] +Hourly 1843 HIGH (`research/notes.md` §129 / composition items +481–496 / findings batch #111). Does **not** bump the 0.5.0 pin. +Uniqueness dumps live in +[`research/changelog-hourly.md`](research/changelog-hourly.md). +Do not reopen or amend PR #23–#51. +Do not amend released 0.5.0 (#42). Merged #51 owns §128. Merged #50 owns §127. + +### Added + +- **Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify / + --init_from warm-start LoRA/head PR #9 / Kev-0.8B 4B 9B Qwen3.5 family / + from-scratch ≠ warm-start / JSONL labels ≠ Harbor / + reconstruction ≠ replica / assay-001 split verdict / + catalogs as indexes. 4B new-source 0.794/0.832 *theirs*. + 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. + wire-compat is not logit-equiv. SHA move is not a replica. + Evaluator: from-scratch ≠ warm-start / JSONL labels ≠ Harbor / + reconstruction ≠ replica. uniqueness_gate.py now + checks 0843 + 0915 + jcr + 0922 + 0940 + 0947 + 1049 + 1143 + + 1248 + 1340 + 1441 + 1542 + 1643 + 1746 + 1843. Composition items 481–496 / + batch #111. + **HARD RULE:** do not reopen or amend PR #23–#51. Does **not** bump + 0.5.0. + +- **Recipe (class, not Jev-only).** Without Augustus: treat a from-scratch + fine-tune as equal to a warm start, a JSONL file as Harbor, a random- + weight reconstruction as a replica, or a catalog as a grant. With + Augustus: from-scratch ≠ warm-start; JSONL labels ≠ Harbor; + reconstruction ≠ replica; catalog ≠ endorsement. Same split for any + Choice/Score/Noul-style head, not only hosted Jev. + Hourly 1746 HIGH (`research/notes.md` §128 / composition items 465–480 / findings batch #110). Does **not** bump the 0.5.0 pin. Uniqueness dumps live in diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 679d322..7ba716f 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -44,8 +44,8 @@ re-opened as "new." Before folding: - Read `research/notes.md` and the uniqueness fragments in `.agents/skills/augustus/SKILL.md` - Do not re-fold an already-landed section as a new beat -- Do not reopen or amend a merged fold PR (#23–#50) -- uniqueness_gate.py checks 0843 + 0915 + jcr + 0922 + 0940 + 0947 + 1049 + 1143 + 1248 + 1340 + 1441 + 1542 + 1643 + 1746 consecutive locks, plus the revisit / since-last-look protocol substring in the skill and research files. +- Do not reopen or amend a merged fold PR (#23–#51) +- uniqueness_gate.py checks 0843 + 0915 + jcr + 0922 + 0940 + 0947 + 1049 + 1143 + 1248 + 1340 + 1441 + 1542 + 1643 + 1746 + 1843 consecutive locks, plus the revisit / since-last-look protocol substring in the skill and research files. - Hourly uniqueness dump: `research/changelog-hourly.md` (archive, not release notes) - Treat **revisit HIGH like novel HIGH**. Catalogued repos are not diff --git a/README.md b/README.md index b9a90db..392d725 100644 --- a/README.md +++ b/README.md @@ -176,3 +176,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/docs/_includes/recipes.html b/docs/_includes/recipes.html index fea40be..e1da63e 100644 --- a/docs/_includes/recipes.html +++ b/docs/_includes/recipes.html @@ -278,6 +278,49 @@

Win rate vs Noul

Sim metrics separately from Brier/ECE. *theirs* not Harbor.
+ +
+

Open-head training

+

From-scratch vs warm-start

+
+
Problem
+
A from-scratch adapter treated as equal to a released head plus your labels.
+
Without
+
Train from the base. Quote 0.88 as Harbor. Skip the held-out file.
+
With
+
from-scratch ≠ warm-start. JSONL labels ≠ Harbor. --init_from keeps what the head already knows.
+
Measure
+
0.33 vs 0.84 vs 0.83/0.88 *theirs*. Same split for any Choice/Score/Noul head.
+
+
+
+

Independent assay

+

Split verdict vs global certificate

+
+
Problem
+
One corpus ECE treated as the model's calibration everywhere.
+
Without
+
Ship CLINC150 as proof. Ignore Banking77 overconfidence.
+
With
+
assay-001 split verdict. A result applies to the artifacts examined.
+
Measure
+
CLINC150 ECE 0.0204 *theirs*. Banking77 ECE 0.0936 *theirs*. Not Harbor.
+
+
+
+

Reconstruction

+

Random weights vs replica

+
+
Problem
+
A from-first-principles sketch treated as TypeSafe logits.
+
Without
+
Swap the serving URL. Quote the catalog as a grant.
+
With
+
reconstruction ≠ replica. unofficial research implementation with random weights. catalog ≠ endorsement.
+
Measure
+
Namesake lock first. Third-party benches stay *theirs*.
+
+

Measurement recipe (hysteresis, equal-width vs quantile ECE, hop-ECE, diff --git a/docs/ecosystem.md b/docs/ecosystem.md index 052d1a0..ca30d61 100644 --- a/docs/ecosystem.md +++ b/docs/ecosystem.md @@ -1163,3 +1163,5 @@ Hourly 1643 uniqueness lock: razorback16/openjev densify HEAD febf02e88989 READM **Hourly 1746 HIGH (`notes.md` §128).** TypeLLM truncated thinking densify. 0.8B thinking On 0/18 *theirs*. forced closure 20/20 type-valid *theirs*. Constrained AR ≠ calibrated Noul. Kev-0.8B completes family. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. transfer-v9 5%/9%/26% *theirs*. Facts go to code. Judgments go to Jev. Only facts can block. Jev never blocks. five-lines threshold 0.80 still soft. 155/155 argmax *theirs*. Qwen3.5 ≠ Archer. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#50. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/research/archive/findings.md b/research/archive/findings.md index 474ce75..8f01a52 100644 --- a/research/archive/findings.md +++ b/research/archive/findings.md @@ -2,6 +2,43 @@ +## Batch #111 (2026-09-20 ~18:43 Boise / ~00:43 UTC) - hourly 1843 HIGH + +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 + +Note: `research/notes.md` §129. Docs + evaluator, fresh PR off latest +`main` (`26915ca` / merged #51 hourly 1746). Merged #51 owns §128. +Merged #50 owns §127. This fold +stays §129 / items 481–496 / batch #111. +**HARD RULE:** do not reopen or amend PR #23–#51. +Quote READMEs. Soft Noul ≠ hard safety. Augustus owns +placement. `invented_signal: false`. + +- **kev own-data densify PRIMARY.** HEAD bd058057ad0a. Fine-tuning on your own data. + --data JSONL. --init_from warm-start LoRA/head PR #9. + Kev-0.8B 4B 9B Qwen3.5 family. 4B new-source 0.794/0.832 *theirs*. + 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. + from-scratch ≠ warm-start. JSONL labels ≠ Harbor. +- **dabit3 / tenbin densify.** Already catalogued. densify is not a sibling. + dabit3/jev-experiments densify 340★. simota/tenbin densify neighbor skill. +- **reconstruction / assay / jev-ra.** reconstruction ≠ replica. + unofficial research implementation with random weights. + assay-001 split verdict. CLINC150 ECE 0.0204 *theirs*. + Banking77 ECE 0.0936 *theirs*. brnyxx/jev-ra 3-5x / ~300 ms *theirs*. +- **catalogs / MCP / HF.** Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev. + ThePFMind/jev-mcp ≠ jkudish/jev-mcp. serving substrate ≠ calibrated replica. + catalog ≠ endorsement. +- **Namesakes / skip.** kyegomez/open-jev ≠ razorback16/openjev. + namenu/pi-jev-effort ≠ TheoOliveira/pi-jev. + samatv256/mini-Jev ≠ r-ms/mini-jev. + comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper. + Archer still promised_not_landed. + +Pulse: Archer still NOT landed. Hub archerhume/4rcherhume HTTP **401**. +kev **1033★**. dabit3 **340★**. Promethe-us/awesome-jev **15★**. +`invented_signal: false`. + + ## Batch #110 (2026-09-20 ~17:46 Boise / ~23:46 UTC) - hourly 1746 HIGH Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 diff --git a/research/archive/hourly/2026-09-21T00/augustus_items.json b/research/archive/hourly/2026-09-21T00/augustus_items.json new file mode 100644 index 0000000..796a90b --- /dev/null +++ b/research/archive/hourly/2026-09-21T00/augustus_items.json @@ -0,0 +1,452 @@ +[ + { + "id": "dabit3/jev-experiments", + "kind": "NOVEL_HIGH", + "description": "TypeSafe / Jev latency-focused demos built by Devin. Each app in its own top-level directory.", + "stars": 340, + "html_url": "https://github.com/dabit3/jev-experiments", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Promethe-us/awesome-jev", + "kind": "NOVEL_HIGH", + "description": "awesome-jev list (no description field)", + "stars": 15, + "html_url": "https://github.com/Promethe-us/awesome-jev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "DragosTana/JEV-FC", + "kind": "NOVEL_HIGH", + "description": "(empty description; name JEV-FC)", + "stars": 2, + "html_url": "https://github.com/DragosTana/JEV-FC", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Nutlope/jev-fraud", + "kind": "NOVEL_HIGH", + "description": "Use Jev with Kimi K3 for hard classification", + "stars": 2, + "html_url": "https://github.com/Nutlope/jev-fraud", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "comoc/jev-minesweeper", + "kind": "NOVEL_HIGH", + "description": "TypeSafe Jev (System One) \u306b\u30d6\u30e9\u30a6\u30b6\u4e0a\u306e\u30de\u30a4\u30f3\u30b9\u30a4\u30fc\u30d1\u30fc\u3092\u89e3\u304b\u305b\u308b\u30c7\u30e2", + "stars": 2, + "html_url": "https://github.com/comoc/jev-minesweeper", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "jeffloo886/jev-notion", + "kind": "NOVEL_HIGH", + "description": "Fill Notion database properties with calibrated AI judgments (TypeSafe Jev). Native macOS SwiftUI + CLI.", + "stars": 2, + "html_url": "https://github.com/jeffloo886/jev-notion", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "simota/tenbin", + "kind": "NOVEL_HIGH", + "description": "MCP server and agent skill for the TypeSafe AI System One API (Jev): decompose a judgment into Choice / Score / Noul questions, lint them, measure on labelled data, and put calibrated thresholds in code", + "stars": 2, + "html_url": "https://github.com/simota/tenbin", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "2456868764/jevguide", + "kind": "NOVEL_HIGH", + "description": "Curated Jev showcases from X, organized by category with media previews and direct source links.", + "stars": 1, + "html_url": "https://github.com/2456868764/jevguide", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "EnesYilmazcode/JevMinesweeper", + "kind": "NOVEL_HIGH", + "description": "Jev Plays Minesweeper", + "stars": 1, + "html_url": "https://github.com/EnesYilmazcode/JevMinesweeper", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Solido/jev_dart", + "kind": "NOVEL_HIGH", + "description": "Typesafe Jev Api", + "stars": 1, + "html_url": "https://github.com/Solido/jev_dart", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "andyhorn/jev", + "kind": "NOVEL_HIGH", + "description": "(empty description; name jev)", + "stars": 1, + "html_url": "https://github.com/andyhorn/jev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "brnyxx/jev-ra", + "kind": "NOVEL_HIGH", + "description": "Browser use for coding agents, 3-5x faster than browser-use. MCP server + CLI; TypeSafe Jev decides every step in ~300 ms.", + "stars": 1, + "html_url": "https://github.com/brnyxx/jev-ra", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:samatv256/mini-Jev", + "kind": "NOVEL_HIGH", + "description": "decision-model system-one", + "stars": 1, + "html_url": "https://huggingface.co/samatv256/mini-Jev", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "kyegomez/open-jev", + "kind": "NOVEL_HIGH", + "description": "an open-source, from-first-principles reconstruction of the ideas behind TypeSafe AI's Jev, written in pytorch", + "stars": 1, + "html_url": "https://github.com/kyegomez/open-jev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "1npo/jev-gmail-labeler", + "kind": "NOVEL_HIGH", + "description": "A tool that uses Jev to label my emails", + "stars": 0, + "html_url": "https://github.com/1npo/jev-gmail-labeler", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Aley3567/awsome-jev-sight", + "kind": "NOVEL_HIGH", + "description": "bright-sight \u7684\u72ec\u7acb\u4f18\u5316\u526f\u672c\uff1a\u628a\u4e00\u53e5\u4e2d\u6587\u6307\u4ee4\u53d8\u6210\u53d7\u63a7\u7684\u5e94\u7528\u64cd\u4f5c\uff0c\u6bcf\u4e00\u6b65\u6267\u884c\u5b8c\u90fd\u7531\u4ee3\u7801\u56de\u8bfb\u9a8c\u8bc1", + "stars": 0, + "html_url": "https://github.com/Aley3567/awsome-jev-sight", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Carl-Lee91/Jev-VoC", + "kind": "NOVEL_HIGH", + "description": "Jev \ud1a0\uc774 \ud504\ub85c\uc81d\ud2b8", + "stars": 0, + "html_url": "https://github.com/Carl-Lee91/Jev-VoC", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "MartinPuli/f1", + "kind": "NOVEL_HIGH", + "description": "JEV Prix: five AI drivers, unknown procedural circuits, Formula-inspired racing, BYOK Jev and saved replays.", + "stars": 0, + "html_url": "https://github.com/MartinPuli/f1", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "SeanPlusPlus/hellojev", + "kind": "NOVEL_HIGH", + "description": "\ud83d\udc4b jev", + "stars": 0, + "html_url": "https://github.com/SeanPlusPlus/hellojev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "SuperInstance/jev-quilt", + "kind": "NOVEL_HIGH", + "description": "JEV for quilt as understood output: cellular-first decision substrate \u2014 typed cells, hook-and-drop deltas, bookkeeper WAL, last-mile projection decoupled", + "stars": 0, + "html_url": "https://github.com/SuperInstance/jev-quilt", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "ThePFMind/jev-mcp", + "kind": "NOVEL_HIGH", + "description": "MCP server exposing TypeSafe AI's Jev decision model to Claude (stdio, two tools: jev_evaluate, jev_route)", + "stars": 0, + "html_url": "https://github.com/ThePFMind/jev-mcp", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "asmirrr/DriftLab", + "kind": "NOVEL_HIGH", + "description": "Reproducible quantitative research CLI for testing momentum strategies and auditing research methodology with TypeSafe Jev.", + "stars": 0, + "html_url": "https://github.com/asmirrr/DriftLab", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "bouncerguy/jevwrapper", + "kind": "NOVEL_HIGH", + "description": "English in. Typed JEV decisions out. Inspectable LLM-to-JEV middleware, a browser sandbox, and reusable examples. MIT licensed.", + "stars": 0, + "html_url": "https://github.com/bouncerguy/jevwrapper", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "cristiancolon/jev-hft", + "kind": "NOVEL_HIGH", + "description": "Research pipeline testing whether TypeSafe's Jev (via Vercel AI Gateway) can judge news and market data fast enough to matter. Bitcoin and US stocks, paper trading only.", + "stars": 0, + "html_url": "https://github.com/cristiancolon/jev-hft", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "digitalfoudnry-vb/JevBrowser", + "kind": "NOVEL_HIGH", + "description": "JEV Browser", + "stars": 0, + "html_url": "https://github.com/digitalfoudnry-vb/JevBrowser", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "dtduc-git/jevnav", + "kind": "NOVEL_HIGH", + "description": "Browser automation whose decisions you can replay, test and audit \u2014 Jev picks the element, traces replay offline in CI", + "stars": 0, + "html_url": "https://github.com/dtduc-git/jevnav", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "gyozameronpanofficial/human-exe-jev", + "kind": "NOVEL_HIGH", + "description": "HUMAN.exe \u2014 Jev-powered psychological checkpoint game", + "stars": 0, + "html_url": "https://github.com/gyozameronpanofficial/human-exe-jev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hawkyre/jevx", + "kind": "NOVEL_HIGH", + "description": "Find relevant conversations on X and score your drafts. Open-source Chrome and Firefox extension powered by Jev.", + "stars": 0, + "html_url": "https://github.com/hawkyre/jevx", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:BlueAquilae/laya-pack-shaped", + "kind": "NOVEL_HIGH", + "description": "safetensors region:us", + "stars": 0, + "html_url": "https://huggingface.co/BlueAquilae/laya-pack-shaped", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:akhilaaa3/openjev-v1-allmix-r512-merged", + "kind": "NOVEL_HIGH", + "description": "safetensors text-classification merged base_model:google/gemma-4-12B-it base_model:finetune:google/gemma-4-12B-it license:apache-2.0 region:us", + "stars": 0, + "html_url": "https://huggingface.co/akhilaaa3/openjev-v1-allmix-r512-merged", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:cklxx/laya-browser", + "kind": "NOVEL_HIGH", + "description": "safetensors laya system-1 browser-agent web-navigation decision-model modernbert mind2web tilelang en multilingual dataset:osunlp/Mind2Web", + "stars": 0, + "html_url": "https://huggingface.co/cklxx/laya-browser", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:frankyy03/laya-pt-es-nli", + "kind": "NOVEL_HIGH", + "description": "laya safetensors mmbert natural-language-inference portuguese spanish calibrated-decisions rlcd audio-adapter-compatible text-classification pt es", + "stars": 0, + "html_url": "https://huggingface.co/frankyy03/laya-pt-es-nli", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "jourdanlabs/assay-001", + "kind": "NOVEL_HIGH", + "description": "ASSAY-001: independent, pre-registered verification of TypeSafe Jev's calibration and type-safety claims. Split verdict, published in full.", + "stars": 0, + "html_url": "https://github.com/jourdanlabs/assay-001", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "kijung4290/maeum-on-attendance-care", + "kind": "NOVEL_HIGH", + "description": "\uc5b4\ub974\uc2e0 \ud504\ub85c\uadf8\ub7a8 \ucd9c\uc11d \uc704\ud5d8 \ubaa8\ub2c8\ud130\ub9c1 \ub300\uc2dc\ubcf4\ub4dc \u00b7 TypeSafe AI JEV \uc5f0\ub3d9", + "stars": 0, + "html_url": "https://github.com/kijung4290/maeum-on-attendance-care", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "lafollett-labs/typesafe-jev-dojo", + "kind": "NOVEL_HIGH", + "description": "A live, graphical dojo for TypeSafe's Jev (System One) typed decision model \u2014 routing, a Tetris-playing agent, parallel swarms, and an honest Jev-vs-Claude gauntlet.", + "stars": 0, + "html_url": "https://github.com/lafollett-labs/typesafe-jev-dojo", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "mateoromano-personal/contact-cleaner", + "kind": "NOVEL_HIGH", + "description": "Sorts the thousands of addresses Google saved automatically into keep, review and remove. Next.js + Jev.", + "stars": 0, + "html_url": "https://github.com/mateoromano-personal/contact-cleaner", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "namenu/pi-jev-effort", + "kind": "NOVEL_HIGH", + "description": "Pi extension: per-prompt thinking level from a TypeSafe Jev judgement, capped by the quota you have left", + "stars": 0, + "html_url": "https://github.com/namenu/pi-jev-effort", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "nanami-0713/dsh-jev-decide", + "kind": "NOVEL_HIGH", + "description": "DSH plugin: register TypeSafe Jev (System One decision model) as an agent tool \u2014 jev_decide returns calibrated probabilities (noul/choice/score) for routing/triage/guardrail judgments, no text generation. \u628a TypeSafe Jev \u51b3\u7b56\u6a21\u578b\u6ce8\u518c\u4e3a DSH agent \u5de5\u5177", + "stars": 0, + "html_url": "https://github.com/nanami-0713/dsh-jev-decide", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "peterrauscher/x-bookmark-sorter", + "kind": "NOVEL_HIGH", + "description": "Chrome extension (Manifest V3) that auto-sorts your X bookmarks into your own folders using TypeSafe's Jev decision model.", + "stars": 0, + "html_url": "https://github.com/peterrauscher/x-bookmark-sorter", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "sindunarasimhan/murmur", + "kind": "NOVEL_HIGH", + "description": "Voice-first podcast listening with Jev and OpenAI.", + "stars": 0, + "html_url": "https://github.com/sindunarasimhan/murmur", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "treycausey/semantic-find", + "kind": "NOVEL_HIGH", + "description": "Local evidence search with optional Jev semantic ranking", + "stars": 0, + "html_url": "https://github.com/treycausey/semantic-find", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "ttlequals0/MinusPodJev", + "kind": "NOVEL_HIGH", + "description": "MinusPod Jev Proxy", + "stars": 0, + "html_url": "https://github.com/ttlequals0/MinusPodJev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "yangzhou-chaofan/awesome-jev-prompt", + "kind": "NOVEL_HIGH", + "description": "latest top 100 showcases for jev (keep updating) from x / github / latest sources", + "stars": 0, + "html_url": "https://github.com/yangzhou-chaofan/awesome-jev-prompt", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "yasuhito/jev-computer-use", + "kind": "NOVEL_HIGH", + "description": "Safe computer-use automation guided by typed Jev decisions", + "stars": 0, + "html_url": "https://github.com/yasuhito/jev-computer-use", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "jaredpalmer/kev", + "kind": "REVISIT_HIGH", + "why": "HEAD bd058057ad0a (was 38087aa0301d): 6 commits \u2014 Fine-tuning on your own data (--data JSONL for kev.train/benchmark), --init_from warm-start LoRA/head (PR #9), Kev-0.5B card Qwen3.5 family pointer, README recipe, tests. Material training-API + README densify *theirs*", + "stars": 1032, + "source": "github", + "note": "densify existing card; mark *theirs*; no equivalence guarantee", + "prior_sha": "38087aa0301d800e12d59bbf848f582a80c207bc", + "new_sha": "bd058057ad0aa9df3dd6d14e3542c95a5ce367b2" + } +] \ No newline at end of file diff --git a/research/archive/hourly/2026-09-21T00/novel_high_this_run.json b/research/archive/hourly/2026-09-21T00/novel_high_this_run.json new file mode 100644 index 0000000..f613278 --- /dev/null +++ b/research/archive/hourly/2026-09-21T00/novel_high_this_run.json @@ -0,0 +1,442 @@ +[ + { + "id": "dabit3/jev-experiments", + "kind": "NOVEL_HIGH", + "description": "TypeSafe / Jev latency-focused demos built by Devin. Each app in its own top-level directory.", + "stars": 340, + "html_url": "https://github.com/dabit3/jev-experiments", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Promethe-us/awesome-jev", + "kind": "NOVEL_HIGH", + "description": "awesome-jev list (no description field)", + "stars": 15, + "html_url": "https://github.com/Promethe-us/awesome-jev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "DragosTana/JEV-FC", + "kind": "NOVEL_HIGH", + "description": "(empty description; name JEV-FC)", + "stars": 2, + "html_url": "https://github.com/DragosTana/JEV-FC", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Nutlope/jev-fraud", + "kind": "NOVEL_HIGH", + "description": "Use Jev with Kimi K3 for hard classification", + "stars": 2, + "html_url": "https://github.com/Nutlope/jev-fraud", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "comoc/jev-minesweeper", + "kind": "NOVEL_HIGH", + "description": "TypeSafe Jev (System One) \u306b\u30d6\u30e9\u30a6\u30b6\u4e0a\u306e\u30de\u30a4\u30f3\u30b9\u30a4\u30fc\u30d1\u30fc\u3092\u89e3\u304b\u305b\u308b\u30c7\u30e2", + "stars": 2, + "html_url": "https://github.com/comoc/jev-minesweeper", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "jeffloo886/jev-notion", + "kind": "NOVEL_HIGH", + "description": "Fill Notion database properties with calibrated AI judgments (TypeSafe Jev). Native macOS SwiftUI + CLI.", + "stars": 2, + "html_url": "https://github.com/jeffloo886/jev-notion", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "simota/tenbin", + "kind": "NOVEL_HIGH", + "description": "MCP server and agent skill for the TypeSafe AI System One API (Jev): decompose a judgment into Choice / Score / Noul questions, lint them, measure on labelled data, and put calibrated thresholds in code", + "stars": 2, + "html_url": "https://github.com/simota/tenbin", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "2456868764/jevguide", + "kind": "NOVEL_HIGH", + "description": "Curated Jev showcases from X, organized by category with media previews and direct source links.", + "stars": 1, + "html_url": "https://github.com/2456868764/jevguide", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "EnesYilmazcode/JevMinesweeper", + "kind": "NOVEL_HIGH", + "description": "Jev Plays Minesweeper", + "stars": 1, + "html_url": "https://github.com/EnesYilmazcode/JevMinesweeper", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Solido/jev_dart", + "kind": "NOVEL_HIGH", + "description": "Typesafe Jev Api", + "stars": 1, + "html_url": "https://github.com/Solido/jev_dart", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "andyhorn/jev", + "kind": "NOVEL_HIGH", + "description": "(empty description; name jev)", + "stars": 1, + "html_url": "https://github.com/andyhorn/jev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "brnyxx/jev-ra", + "kind": "NOVEL_HIGH", + "description": "Browser use for coding agents, 3-5x faster than browser-use. MCP server + CLI; TypeSafe Jev decides every step in ~300 ms.", + "stars": 1, + "html_url": "https://github.com/brnyxx/jev-ra", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:samatv256/mini-Jev", + "kind": "NOVEL_HIGH", + "description": "decision-model system-one", + "stars": 1, + "html_url": "https://huggingface.co/samatv256/mini-Jev", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "kyegomez/open-jev", + "kind": "NOVEL_HIGH", + "description": "an open-source, from-first-principles reconstruction of the ideas behind TypeSafe AI's Jev, written in pytorch", + "stars": 1, + "html_url": "https://github.com/kyegomez/open-jev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "1npo/jev-gmail-labeler", + "kind": "NOVEL_HIGH", + "description": "A tool that uses Jev to label my emails", + "stars": 0, + "html_url": "https://github.com/1npo/jev-gmail-labeler", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Aley3567/awsome-jev-sight", + "kind": "NOVEL_HIGH", + "description": "bright-sight \u7684\u72ec\u7acb\u4f18\u5316\u526f\u672c\uff1a\u628a\u4e00\u53e5\u4e2d\u6587\u6307\u4ee4\u53d8\u6210\u53d7\u63a7\u7684\u5e94\u7528\u64cd\u4f5c\uff0c\u6bcf\u4e00\u6b65\u6267\u884c\u5b8c\u90fd\u7531\u4ee3\u7801\u56de\u8bfb\u9a8c\u8bc1", + "stars": 0, + "html_url": "https://github.com/Aley3567/awsome-jev-sight", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "Carl-Lee91/Jev-VoC", + "kind": "NOVEL_HIGH", + "description": "Jev \ud1a0\uc774 \ud504\ub85c\uc81d\ud2b8", + "stars": 0, + "html_url": "https://github.com/Carl-Lee91/Jev-VoC", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "MartinPuli/f1", + "kind": "NOVEL_HIGH", + "description": "JEV Prix: five AI drivers, unknown procedural circuits, Formula-inspired racing, BYOK Jev and saved replays.", + "stars": 0, + "html_url": "https://github.com/MartinPuli/f1", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "SeanPlusPlus/hellojev", + "kind": "NOVEL_HIGH", + "description": "\ud83d\udc4b jev", + "stars": 0, + "html_url": "https://github.com/SeanPlusPlus/hellojev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "SuperInstance/jev-quilt", + "kind": "NOVEL_HIGH", + "description": "JEV for quilt as understood output: cellular-first decision substrate \u2014 typed cells, hook-and-drop deltas, bookkeeper WAL, last-mile projection decoupled", + "stars": 0, + "html_url": "https://github.com/SuperInstance/jev-quilt", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "ThePFMind/jev-mcp", + "kind": "NOVEL_HIGH", + "description": "MCP server exposing TypeSafe AI's Jev decision model to Claude (stdio, two tools: jev_evaluate, jev_route)", + "stars": 0, + "html_url": "https://github.com/ThePFMind/jev-mcp", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "asmirrr/DriftLab", + "kind": "NOVEL_HIGH", + "description": "Reproducible quantitative research CLI for testing momentum strategies and auditing research methodology with TypeSafe Jev.", + "stars": 0, + "html_url": "https://github.com/asmirrr/DriftLab", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "bouncerguy/jevwrapper", + "kind": "NOVEL_HIGH", + "description": "English in. Typed JEV decisions out. Inspectable LLM-to-JEV middleware, a browser sandbox, and reusable examples. MIT licensed.", + "stars": 0, + "html_url": "https://github.com/bouncerguy/jevwrapper", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "cristiancolon/jev-hft", + "kind": "NOVEL_HIGH", + "description": "Research pipeline testing whether TypeSafe's Jev (via Vercel AI Gateway) can judge news and market data fast enough to matter. Bitcoin and US stocks, paper trading only.", + "stars": 0, + "html_url": "https://github.com/cristiancolon/jev-hft", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "digitalfoudnry-vb/JevBrowser", + "kind": "NOVEL_HIGH", + "description": "JEV Browser", + "stars": 0, + "html_url": "https://github.com/digitalfoudnry-vb/JevBrowser", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "dtduc-git/jevnav", + "kind": "NOVEL_HIGH", + "description": "Browser automation whose decisions you can replay, test and audit \u2014 Jev picks the element, traces replay offline in CI", + "stars": 0, + "html_url": "https://github.com/dtduc-git/jevnav", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "gyozameronpanofficial/human-exe-jev", + "kind": "NOVEL_HIGH", + "description": "HUMAN.exe \u2014 Jev-powered psychological checkpoint game", + "stars": 0, + "html_url": "https://github.com/gyozameronpanofficial/human-exe-jev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hawkyre/jevx", + "kind": "NOVEL_HIGH", + "description": "Find relevant conversations on X and score your drafts. Open-source Chrome and Firefox extension powered by Jev.", + "stars": 0, + "html_url": "https://github.com/hawkyre/jevx", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:BlueAquilae/laya-pack-shaped", + "kind": "NOVEL_HIGH", + "description": "safetensors region:us", + "stars": 0, + "html_url": "https://huggingface.co/BlueAquilae/laya-pack-shaped", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:akhilaaa3/openjev-v1-allmix-r512-merged", + "kind": "NOVEL_HIGH", + "description": "safetensors text-classification merged base_model:google/gemma-4-12B-it base_model:finetune:google/gemma-4-12B-it license:apache-2.0 region:us", + "stars": 0, + "html_url": "https://huggingface.co/akhilaaa3/openjev-v1-allmix-r512-merged", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:cklxx/laya-browser", + "kind": "NOVEL_HIGH", + "description": "safetensors laya system-1 browser-agent web-navigation decision-model modernbert mind2web tilelang en multilingual dataset:osunlp/Mind2Web", + "stars": 0, + "html_url": "https://huggingface.co/cklxx/laya-browser", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "hf:frankyy03/laya-pt-es-nli", + "kind": "NOVEL_HIGH", + "description": "laya safetensors mmbert natural-language-inference portuguese spanish calibrated-decisions rlcd audio-adapter-compatible text-classification pt es", + "stars": 0, + "html_url": "https://huggingface.co/frankyy03/laya-pt-es-nli", + "source": "hf_model", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "jourdanlabs/assay-001", + "kind": "NOVEL_HIGH", + "description": "ASSAY-001: independent, pre-registered verification of TypeSafe Jev's calibration and type-safety claims. Split verdict, published in full.", + "stars": 0, + "html_url": "https://github.com/jourdanlabs/assay-001", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "kijung4290/maeum-on-attendance-care", + "kind": "NOVEL_HIGH", + "description": "\uc5b4\ub974\uc2e0 \ud504\ub85c\uadf8\ub7a8 \ucd9c\uc11d \uc704\ud5d8 \ubaa8\ub2c8\ud130\ub9c1 \ub300\uc2dc\ubcf4\ub4dc \u00b7 TypeSafe AI JEV \uc5f0\ub3d9", + "stars": 0, + "html_url": "https://github.com/kijung4290/maeum-on-attendance-care", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "lafollett-labs/typesafe-jev-dojo", + "kind": "NOVEL_HIGH", + "description": "A live, graphical dojo for TypeSafe's Jev (System One) typed decision model \u2014 routing, a Tetris-playing agent, parallel swarms, and an honest Jev-vs-Claude gauntlet.", + "stars": 0, + "html_url": "https://github.com/lafollett-labs/typesafe-jev-dojo", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "mateoromano-personal/contact-cleaner", + "kind": "NOVEL_HIGH", + "description": "Sorts the thousands of addresses Google saved automatically into keep, review and remove. Next.js + Jev.", + "stars": 0, + "html_url": "https://github.com/mateoromano-personal/contact-cleaner", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "namenu/pi-jev-effort", + "kind": "NOVEL_HIGH", + "description": "Pi extension: per-prompt thinking level from a TypeSafe Jev judgement, capped by the quota you have left", + "stars": 0, + "html_url": "https://github.com/namenu/pi-jev-effort", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "nanami-0713/dsh-jev-decide", + "kind": "NOVEL_HIGH", + "description": "DSH plugin: register TypeSafe Jev (System One decision model) as an agent tool \u2014 jev_decide returns calibrated probabilities (noul/choice/score) for routing/triage/guardrail judgments, no text generation. \u628a TypeSafe Jev \u51b3\u7b56\u6a21\u578b\u6ce8\u518c\u4e3a DSH agent \u5de5\u5177", + "stars": 0, + "html_url": "https://github.com/nanami-0713/dsh-jev-decide", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "peterrauscher/x-bookmark-sorter", + "kind": "NOVEL_HIGH", + "description": "Chrome extension (Manifest V3) that auto-sorts your X bookmarks into your own folders using TypeSafe's Jev decision model.", + "stars": 0, + "html_url": "https://github.com/peterrauscher/x-bookmark-sorter", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "sindunarasimhan/murmur", + "kind": "NOVEL_HIGH", + "description": "Voice-first podcast listening with Jev and OpenAI.", + "stars": 0, + "html_url": "https://github.com/sindunarasimhan/murmur", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "treycausey/semantic-find", + "kind": "NOVEL_HIGH", + "description": "Local evidence search with optional Jev semantic ranking", + "stars": 0, + "html_url": "https://github.com/treycausey/semantic-find", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "ttlequals0/MinusPodJev", + "kind": "NOVEL_HIGH", + "description": "MinusPod Jev Proxy", + "stars": 0, + "html_url": "https://github.com/ttlequals0/MinusPodJev", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "yangzhou-chaofan/awesome-jev-prompt", + "kind": "NOVEL_HIGH", + "description": "latest top 100 showcases for jev (keep updating) from x / github / latest sources", + "stars": 0, + "html_url": "https://github.com/yangzhou-chaofan/awesome-jev-prompt", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + }, + { + "id": "yasuhito/jev-computer-use", + "kind": "NOVEL_HIGH", + "description": "Safe computer-use automation guided by typed Jev decisions", + "stars": 0, + "html_url": "https://github.com/yasuhito/jev-computer-use", + "source": "github", + "why": "first-seen this run; class-relevant decision/System One tooling or model", + "note": "first-sighting card; mark third-party claims *theirs*" + } +] diff --git a/research/archive/hourly/2026-09-21T00/prompt_Augustus.md b/research/archive/hourly/2026-09-21T00/prompt_Augustus.md new file mode 100644 index 0000000..fe4e16a --- /dev/null +++ b/research/archive/hourly/2026-09-21T00/prompt_Augustus.md @@ -0,0 +1,334 @@ +# Fold hourly System One watch 1843 → design + +Boise label **1843** (~2026-09-20 18:43 MDT). Mode: **NEW_OFF_MAIN**. +Prior: agent bc-84a74271 finished; Augustus #51 still OPEN from 1746 — HARD rule: finished agent → NEW_OFF_MAIN (do not reply into finished agent). + +Role: design judgment / measurement / class table / recipes / densify cards — Augustus-primary for REVISIT; backend-agnostic class (not Jev-only) + +Standing orders: +- Prefer human-facing prose without em dashes or other AI tells (Simon Willison list). +- Mark third-party benches and claims *theirs*; no equivalence guarantees. +- Densify existing catalog cards on REVISIT; first-sighting cards on NOVEL. +- Do not invent metrics. Keep diffs focused on the fold cards / class tables / recipes the repo already uses. +- Open one PR from main with a clear title like `Fold hourly 1843 HIGH`. + +## Items (45) + +### 1. `dabit3/jev-experiments` — NOVEL_HIGH +- Stars/likes context: 340 +- Description: TypeSafe / Jev latency-focused demos built by Devin. Each app in its own top-level directory. +- URL: https://github.com/dabit3/jev-experiments +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 2. `Promethe-us/awesome-jev` — NOVEL_HIGH +- Stars/likes context: 15 +- Description: awesome-jev list (no description field) +- URL: https://github.com/Promethe-us/awesome-jev +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 3. `DragosTana/JEV-FC` — NOVEL_HIGH +- Stars/likes context: 2 +- Description: (empty description; name JEV-FC) +- URL: https://github.com/DragosTana/JEV-FC +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 4. `Nutlope/jev-fraud` — NOVEL_HIGH +- Stars/likes context: 2 +- Description: Use Jev with Kimi K3 for hard classification +- URL: https://github.com/Nutlope/jev-fraud +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 5. `comoc/jev-minesweeper` — NOVEL_HIGH +- Stars/likes context: 2 +- Description: TypeSafe Jev (System One) にブラウザ上のマインスイーパーを解かせるデモ +- URL: https://github.com/comoc/jev-minesweeper +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 6. `jeffloo886/jev-notion` — NOVEL_HIGH +- Stars/likes context: 2 +- Description: Fill Notion database properties with calibrated AI judgments (TypeSafe Jev). Native macOS SwiftUI + CLI. +- URL: https://github.com/jeffloo886/jev-notion +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 7. `simota/tenbin` — NOVEL_HIGH +- Stars/likes context: 2 +- Description: MCP server and agent skill for the TypeSafe AI System One API (Jev): decompose a judgment into Choice / Score / Noul questions, lint them, measure on labelled data, and put calibrated thresholds in code +- URL: https://github.com/simota/tenbin +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 8. `2456868764/jevguide` — NOVEL_HIGH +- Stars/likes context: 1 +- Description: Curated Jev showcases from X, organized by category with media previews and direct source links. +- URL: https://github.com/2456868764/jevguide +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 9. `EnesYilmazcode/JevMinesweeper` — NOVEL_HIGH +- Stars/likes context: 1 +- Description: Jev Plays Minesweeper +- URL: https://github.com/EnesYilmazcode/JevMinesweeper +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 10. `Solido/jev_dart` — NOVEL_HIGH +- Stars/likes context: 1 +- Description: Typesafe Jev Api +- URL: https://github.com/Solido/jev_dart +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 11. `andyhorn/jev` — NOVEL_HIGH +- Stars/likes context: 1 +- Description: (empty description; name jev) +- URL: https://github.com/andyhorn/jev +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 12. `brnyxx/jev-ra` — NOVEL_HIGH +- Stars/likes context: 1 +- Description: Browser use for coding agents, 3-5x faster than browser-use. MCP server + CLI; TypeSafe Jev decides every step in ~300 ms. +- URL: https://github.com/brnyxx/jev-ra +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 13. `hf:samatv256/mini-Jev` — NOVEL_HIGH +- Stars/likes context: 1 +- Description: decision-model system-one +- URL: https://huggingface.co/samatv256/mini-Jev +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 14. `kyegomez/open-jev` — NOVEL_HIGH +- Stars/likes context: 1 +- Description: an open-source, from-first-principles reconstruction of the ideas behind TypeSafe AI's Jev, written in pytorch +- URL: https://github.com/kyegomez/open-jev +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 15. `1npo/jev-gmail-labeler` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: A tool that uses Jev to label my emails +- URL: https://github.com/1npo/jev-gmail-labeler +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 16. `Aley3567/awsome-jev-sight` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: bright-sight 的独立优化副本：把一句中文指令变成受控的应用操作，每一步执行完都由代码回读验证 +- URL: https://github.com/Aley3567/awsome-jev-sight +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 17. `Carl-Lee91/Jev-VoC` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Jev 토이 프로젝트 +- URL: https://github.com/Carl-Lee91/Jev-VoC +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 18. `MartinPuli/f1` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: JEV Prix: five AI drivers, unknown procedural circuits, Formula-inspired racing, BYOK Jev and saved replays. +- URL: https://github.com/MartinPuli/f1 +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 19. `SeanPlusPlus/hellojev` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: 👋 jev +- URL: https://github.com/SeanPlusPlus/hellojev +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 20. `SuperInstance/jev-quilt` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: JEV for quilt as understood output: cellular-first decision substrate — typed cells, hook-and-drop deltas, bookkeeper WAL, last-mile projection decoupled +- URL: https://github.com/SuperInstance/jev-quilt +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 21. `ThePFMind/jev-mcp` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: MCP server exposing TypeSafe AI's Jev decision model to Claude (stdio, two tools: jev_evaluate, jev_route) +- URL: https://github.com/ThePFMind/jev-mcp +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 22. `asmirrr/DriftLab` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Reproducible quantitative research CLI for testing momentum strategies and auditing research methodology with TypeSafe Jev. +- URL: https://github.com/asmirrr/DriftLab +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 23. `bouncerguy/jevwrapper` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: English in. Typed JEV decisions out. Inspectable LLM-to-JEV middleware, a browser sandbox, and reusable examples. MIT licensed. +- URL: https://github.com/bouncerguy/jevwrapper +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 24. `cristiancolon/jev-hft` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Research pipeline testing whether TypeSafe's Jev (via Vercel AI Gateway) can judge news and market data fast enough to matter. Bitcoin and US stocks, paper trading only. +- URL: https://github.com/cristiancolon/jev-hft +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 25. `digitalfoudnry-vb/JevBrowser` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: JEV Browser +- URL: https://github.com/digitalfoudnry-vb/JevBrowser +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 26. `dtduc-git/jevnav` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Browser automation whose decisions you can replay, test and audit — Jev picks the element, traces replay offline in CI +- URL: https://github.com/dtduc-git/jevnav +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 27. `gyozameronpanofficial/human-exe-jev` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: HUMAN.exe — Jev-powered psychological checkpoint game +- URL: https://github.com/gyozameronpanofficial/human-exe-jev +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 28. `hawkyre/jevx` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Find relevant conversations on X and score your drafts. Open-source Chrome and Firefox extension powered by Jev. +- URL: https://github.com/hawkyre/jevx +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 29. `hf:BlueAquilae/laya-pack-shaped` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: safetensors region:us +- URL: https://huggingface.co/BlueAquilae/laya-pack-shaped +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 30. `hf:akhilaaa3/openjev-v1-allmix-r512-merged` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: safetensors text-classification merged base_model:google/gemma-4-12B-it base_model:finetune:google/gemma-4-12B-it license:apache-2.0 region:us +- URL: https://huggingface.co/akhilaaa3/openjev-v1-allmix-r512-merged +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 31. `hf:cklxx/laya-browser` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: safetensors laya system-1 browser-agent web-navigation decision-model modernbert mind2web tilelang en multilingual dataset:osunlp/Mind2Web +- URL: https://huggingface.co/cklxx/laya-browser +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 32. `hf:frankyy03/laya-pt-es-nli` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: laya safetensors mmbert natural-language-inference portuguese spanish calibrated-decisions rlcd audio-adapter-compatible text-classification pt es +- URL: https://huggingface.co/frankyy03/laya-pt-es-nli +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 33. `jourdanlabs/assay-001` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: ASSAY-001: independent, pre-registered verification of TypeSafe Jev's calibration and type-safety claims. Split verdict, published in full. +- URL: https://github.com/jourdanlabs/assay-001 +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 34. `kijung4290/maeum-on-attendance-care` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: 어르신 프로그램 출석 위험 모니터링 대시보드 · TypeSafe AI JEV 연동 +- URL: https://github.com/kijung4290/maeum-on-attendance-care +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 35. `lafollett-labs/typesafe-jev-dojo` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: A live, graphical dojo for TypeSafe's Jev (System One) typed decision model — routing, a Tetris-playing agent, parallel swarms, and an honest Jev-vs-Claude gauntlet. +- URL: https://github.com/lafollett-labs/typesafe-jev-dojo +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 36. `mateoromano-personal/contact-cleaner` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Sorts the thousands of addresses Google saved automatically into keep, review and remove. Next.js + Jev. +- URL: https://github.com/mateoromano-personal/contact-cleaner +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 37. `namenu/pi-jev-effort` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Pi extension: per-prompt thinking level from a TypeSafe Jev judgement, capped by the quota you have left +- URL: https://github.com/namenu/pi-jev-effort +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 38. `nanami-0713/dsh-jev-decide` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: DSH plugin: register TypeSafe Jev (System One decision model) as an agent tool — jev_decide returns calibrated probabilities (noul/choice/score) for routing/triage/guardrail judgments, no text generation. 把 TypeSafe Jev 决策模型注册为 DSH agent 工具 +- URL: https://github.com/nanami-0713/dsh-jev-decide +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 39. `peterrauscher/x-bookmark-sorter` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Chrome extension (Manifest V3) that auto-sorts your X bookmarks into your own folders using TypeSafe's Jev decision model. +- URL: https://github.com/peterrauscher/x-bookmark-sorter +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 40. `sindunarasimhan/murmur` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Voice-first podcast listening with Jev and OpenAI. +- URL: https://github.com/sindunarasimhan/murmur +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 41. `treycausey/semantic-find` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Local evidence search with optional Jev semantic ranking +- URL: https://github.com/treycausey/semantic-find +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 42. `ttlequals0/MinusPodJev` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: MinusPod Jev Proxy +- URL: https://github.com/ttlequals0/MinusPodJev +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 43. `yangzhou-chaofan/awesome-jev-prompt` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: latest top 100 showcases for jev (keep updating) from x / github / latest sources +- URL: https://github.com/yangzhou-chaofan/awesome-jev-prompt +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 44. `yasuhito/jev-computer-use` — NOVEL_HIGH +- Stars/likes context: 0 +- Description: Safe computer-use automation guided by typed Jev decisions +- URL: https://github.com/yasuhito/jev-computer-use +- Why: first-seen this run; class-relevant decision/System One tooling or model +- Note: first-sighting card; mark third-party claims *theirs* + +### 45. `jaredpalmer/kev` — REVISIT_HIGH +- Stars/likes context: 1032 +- Why: HEAD bd058057ad0a (was 38087aa0301d): 6 commits — Fine-tuning on your own data (--data JSONL for kev.train/benchmark), --init_from warm-start LoRA/head (PR #9), Kev-0.5B card Qwen3.5 family pointer, README recipe, tests. Material training-API + README densify *theirs* +- Prior SHA: `38087aa0301d800e12d59bbf848f582a80c207bc` → new `bd058057ad0aa9df3dd6d14e3542c95a5ce367b2` +- Note: densify existing card; mark *theirs*; no equivalence guarantee + +## Success criteria +- Cards/entries folded for each item above (densify REVISIT; first-sight NOVEL). +- PR opened against main. +- No em dashes in human-facing prose you add. \ No newline at end of file diff --git a/research/archive/hourly/2026-09-21T00/revisit_high_this_run.json b/research/archive/hourly/2026-09-21T00/revisit_high_this_run.json new file mode 100644 index 0000000..d1e9e91 --- /dev/null +++ b/research/archive/hourly/2026-09-21T00/revisit_high_this_run.json @@ -0,0 +1,12 @@ +[ + { + "id": "jaredpalmer/kev", + "kind": "REVISIT_HIGH", + "why": "HEAD bd058057ad0a (was 38087aa0301d): 6 commits \u2014 Fine-tuning on your own data (--data JSONL for kev.train/benchmark), --init_from warm-start LoRA/head (PR #9), Kev-0.5B card Qwen3.5 family pointer, README recipe, tests. Material training-API + README densify *theirs*", + "stars": 1032, + "source": "github", + "note": "densify existing card; mark *theirs*; no equivalence guarantee", + "prior_sha": "38087aa0301d800e12d59bbf848f582a80c207bc", + "new_sha": "bd058057ad0aa9df3dd6d14e3542c95a5ce367b2" + } +] diff --git a/research/archive/hourly/2026-09-21T00/run_digest.json b/research/archive/hourly/2026-09-21T00/run_digest.json new file mode 100644 index 0000000..56057f0 --- /dev/null +++ b/research/archive/hourly/2026-09-21T00/run_digest.json @@ -0,0 +1,11 @@ +{ + "hour": "2026-09-21T00", + "label": "1843", + "notes_section": "129", + "composition": "481-496", + "findings_batch": 111, + "revisit_high": 1, + "novel_high": 44, + "invented_signal": false, + "primary": "kev own-data JSONL densify; from-scratch \u2260 warm-start; JSONL labels \u2260 Harbor; reconstruction \u2260 replica" +} diff --git a/research/changelog-hourly.md b/research/changelog-hourly.md index 5a5d99b..f19e3ed 100644 --- a/research/changelog-hourly.md +++ b/research/changelog-hourly.md @@ -17,6 +17,18 @@ This is the uniqueness-lock archive after hourly folds (#2–#40 / notes +## Hourly 1843 HIGH (notes.md §129 / items 481–496 / batch #111) + +- Fresh PR off latest `main` after merged #51 (hourly 1746 / §128) and + merged #50 (hourly 1643 / §127). **HARD RULE:** do not reopen or amend + PR #23–#51. Does not bump 0.5.0. Skip Archer. + Quote *theirs*. No wrappers. `invented_signal: false`. +- kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. + from-scratch ≠ warm-start. JSONL labels ≠ Harbor. + reconstruction ≠ replica. assay-001 split verdict. + SHA move is not a replica. +- Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 + ## Hourly 1746 HIGH (notes.md §128 / items 465–480 / batch #110) - Fresh PR off latest `main` after merged #50 (hourly 1643 / §127) and diff --git a/research/notes.md b/research/notes.md index 2d12fa1..9376fbd 100644 --- a/research/notes.md +++ b/research/notes.md @@ -2505,6 +2505,22 @@ SHA move is not a replica. Full card: `notes.md` §128. +### Since last look (2026-09-21T00 hourly 1843) — jaredpalmer/kev + +DENSIFY §45. Keep this section id. Do not mint a sibling first sighting. +HEAD `bd058057ad0a` README SHA `84b872488915` (was `8465c4c4c294` / 1746). +Quote *theirs*: Fine-tuning on your own data. `--data` JSONL for +`kev.train` / `kev.benchmark`. `--init_from` warm-start LoRA/head (PR #9). +Kev-0.8B 4B 9B Qwen3.5 family. 4B new-source 0.794/0.832 *theirs*. +9B new-source 0.812/0.837 *theirs*. Jev hosted 0.857 *theirs*. +From-scratch on 836 support-tool decisions scored 0.33 against 0.84 for +the released model; `--init_from` kept 0.83 there and reached 0.88 on +the new domain *theirs*. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. +Kev-0.5B card now points at the Qwen3.5 family. No Jev outputs were used +for training. option order can change an answer. +8.2% ≥0.9 on wrong *theirs* (Kev-9B 7.5%). Qwen3.5 ≠ Archer. +SHA move is not a replica. Full card: `notes.md` §129. + ## 46. 14:03 Boise hourly — open multimodal RLCD, bake-off substrate, decision-token LoRA (2026-09-18) America/Boise 14:03 = 20:03 UTC. Docs-only fold into PR #2 @@ -32211,3 +32227,290 @@ Hooks for the reviewer: No live Jev key. No wrappers. Hourly 1746 uniqueness lock: TypeLLM/TypeLLM densify HEAD 702e6a287f3c README SHA 08180db0450b; truncated thinking then constrained decode; typellm_runtime.py typellm_sglang.py; evals/qwen35_small; 0.8B thinking On 0/18 *theirs*; forced closure 20/20 type-valid *theirs*; Constrained AR ≠ calibrated Noul; type safety does not guarantee factual accuracy; Qwen/Qwen3.8-27B ≠ Archer; jaredpalmer/kev densify live HEAD 8465c4c4c294 watch 38087aa0301d README SHA 19664b9ae546; Kev-0.8B completes family; Kev-0.8B 4B 9B Qwen3.5; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; transfer-v9 Kev-9B 5% Jev 9% Kev-8B 26% *theirs*; SemIf Kev-9B 0.917 Jev 0.965 *theirs*; scienthoon 0.952/0.911 vs 0.897/0.914 *theirs*; transformers >= 5.17; Qwen3.5 ≠ Archer; wire-compat ≠ logit-equiv; SHA move is not a replica; notque/vexjoy-agent 421★ /d routes /do fallback; tamaratran/jev-pruner densify HEAD 47d017c34eab; qkal/Canny Facts go to code. Judgments go to Jev. Only facts can block.; Jev never blocks; jqueryscript/awesome-jev 231 entries catalog ≠ endorsement; jqueryscript/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; jamescazzetta/five-lines threshold 0.80 still soft; eugeniughelbur/jev-engineering 371ms $0.0000189 300-call *theirs*; tpellet/jevify ≠ altryne/jevify; seb4ez/jevguard-mcp ≠ seb4ez/jevguard; resumocast/jev-mcp ≠ jkudish/jev-mcp; dtduc-git/jev-table first sighting; Adrian-Ernesto/jevsort ≠ zzzzzec/jevsort; MidasMulli/kev-ane 155/155 argmax *theirs*; MidasMulli/kev-ane ≠ jaredpalmer/kev; serving substrate ≠ calibrated replica; empty repo skip-thin; loktar00/llm-lan-party empty repo; rh-guard owns primary gates; catalog ≠ endorsement; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50; notes.md §128 + +**Hourly 1843 HIGH (`notes.md` §129).** kev own-data JSONL densify. --init_from warm-start LoRA/head PR #9. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. 4B new-source 0.794/0.832 *theirs*. 9B new-source 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. reconstruction ≠ replica. assay-001 split verdict. catalog ≠ endorsement. SHA move is not a replica. Do not reopen or amend PR #23–#51. Does not bump 0.5.0. Skip Archer. `invented_signal: false`. +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 + +## 129. Hourly 1843 HIGH (2026-09-20 ~18:43 Boise / 2026-09-21T00:43Z) + +Measurement / class-member fold on a **fresh PR off latest `main`** +(`cursor/fold-hourly-1843-high-737b`) after `26915ca` (merged #51 +hourly 1746, `notes.md` §128 / items 465–480 / batch #110; merged #50 +hourly 1643, `notes.md` §127). **HARD RULE:** do not reopen or +amend PR #23–#51. Do **not** re-fold §128 1746 / §127 1643 / §126 1542 *as a second census*. +This fold's IDs: `notes.md` §129 / composition 481–496 / +findings batch #111. + +Never reopen merged #7–**#51**. Densify `jaredpalmer/kev` (§45). +`dabit3/jev-experiments` and `simota/tenbin` are already catalogued; +densify, do not mint sibling first-sighting sections. Skip Archer +rewrite. Quote READMEs. Mark *theirs*. No wrappers, keys, `npm` / +`pip` / `uv` / `docker` install recipes. `invented_signal: false`. +Hunches labeled. + +Lane is Augustus: **mathematical / logical / algorithmic mental models** +for Jev-class categorization/scoring across AI / SWE / **business / +knowledge work / life**, not SWE-only. PRIMARY this hour is **REVISIT +densify**: jaredpalmer/kev own-data JSONL fine-tune and `--init_from` +warm-start LoRA/head (PR #9), Kev-0.8B/4B/9B Qwen3.5 family. Class claim +is backend-agnostic: own-data labels plus a warm-started open head is +not a Jev-only trick. from-scratch ≠ warm-start. JSONL labels ≠ Harbor. +Novel HIGH: independent assay split verdict, from-first-principles +reconstruction with random weights, catalogs that are indexes, browser +RA latency *theirs*. Third-party benches stay *theirs*. Wire-compat is +still not logit-equiv. SHA move is not a replica. Catalogs are indexes. +Soft scores ≠ hard gates. reconstruction ≠ replica. Qwen3.5 ≠ Archer. +Archer still **promised_not_landed**. + +Unique consecutive fragments (this hour) must appear as **one +substring** in overlays (see uniqueness gate): +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 + +### How-to-apply (five placements / measurement lenses) + +These are *class* lenses, not vendor tutorials. Same discipline as +§126 (Constrained AR ≠ calibrated Noul) and §127 (restructured vLLM +head ≠ logit-equiv). Formal methods **compose** with scoring: a Noul +is a SENSOR; a JSONL file is not Harbor; a reconstruction with random +weights is not a replica; a catalog is an index. + +1. **from-scratch ≠ warm-start** + (*theirs*, jaredpalmer/kev PRIMARY densify). Quote *theirs*: + Fine-tuning on your own data. Put examples in a JSONL file, one + request per line, with a `label` on every question. `--init_from` + loads the adapter and pointer head from the released model before + training. Starting from the base model instead throws that away: + in one user's test on 836 support-tool decisions, a fine-tune from + the base scored 0.33 on Kev's own evaluation set, against 0.84 for + the released model; the same data with `--init_from` kept 0.83 + there and reached 0.88 on the new domain. from-scratch ≠ warm-start. + This is class-wide for any open Choice/Score/Noul head, not a + Jev-only recipe. Do **not** copy `uv` / Modal / ports. +2. **JSONL labels ≠ Harbor** + (kev.train / kev.benchmark). Quote *theirs*: the same shape as an + API request, plus a label. Keep 10–20% aside for evaluation. The + benchmark reports accuracy, Brier, and calibration per question + type. Those numbers are the author's labelled file, not a frozen + Harbor protocol. JSONL labels ≠ Harbor. SHA move is not a replica. + Life analogue: grading your own homework is not an independent exam. +3. **reconstruction ≠ replica** + (kyegomez/open-jev). Quote *theirs*: an open-source, from-first- + principles reconstruction of the ideas behind TypeSafe AI's Jev, + written in pytorch. Unofficial research implementation with random + weights. It is not the production Jev model, does not reproduce + TypeSafe AI's training data or weights, and makes no claim of + matching their published results. reconstruction ≠ replica. + kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ + Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev. wire-compat ≠ logit-equiv. +4. **assay-001 split verdict** + (jourdanlabs/assay-001). Quote *theirs*: on CLINC150, Jev's chosen- + option probabilities were calibrated (ECE 0.0204); on Banking77 + they were not (ECE 0.0936, systematically overconfident). Across + 8,576 responses there were zero type errors. A result applies to + the artifacts and criteria examined. It is not a statement about + Jev on any other task, corpus, or day. Split verdict is not a + global calibration certificate. *theirs* not Harbor. +5. **catalog ≠ endorsement / game success ≠ calibrated Noul** + (Promethe-us/awesome-jev; minesweeper cousins; brnyxx/jev-ra). + Quote *theirs* awesome-jev: a curated list, community compiled, + not affiliated with TypeSafe. Promethe-us/awesome-jev ≠ + MrJev/awesome-jev ≠ yibie/awesome-jev. Quote *theirs* jev-ra: + 3-5x faster than browser-use; TypeSafe Jev decides every step in + ~300 ms; Wikipedia 8.50× *theirs*. Latency is a systems comparison, + not ECE. comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper. + game success ≠ calibrated Noul. + +### HIGH (revisit densify; keep original section ids) + +1. **[jaredpalmer/kev](https://github.com/jaredpalmer/kev)** + - DENSIFY §45 (Apache-2.0; **1033★**; HEAD `bd058057ad0a`; + README SHA `84b872488915`; was `b339f446a0ef` / 1542 densify). + Fine-tuning on your own data. --data JSONL. + --init_from warm-start LoRA/head PR #9. Kev-0.8B 4B 9B Qwen3.5 + family. 4B new-source 0.794/0.832 *theirs*. 9B new-source + 0.812/0.837 *theirs*. 0.33 vs 0.84 vs 0.83/0.88 *theirs*. + from-scratch ≠ warm-start. JSONL labels ≠ Harbor. Kev-0.5B card + Qwen3.5 family pointer. No Jev outputs were used for training. + option order can change an answer. 8.2% ≥0.9 on wrong *theirs*. + Kev-9B 7.5% ≥0.9 on wrong *theirs*. Qwen3.5 ≠ Archer. SHA move + is not a replica. training-API densify *theirs*. Do **not** + copy `uv` / Modal / keys. + +2. **[dabit3/jev-experiments](https://github.com/dabit3/jev-experiments)** + - DENSIFY the latency-first archive card (license null; **340★**; + HEAD `c469e5bfdc73`; README SHA `ce01b8ab7f38`). Already in the + census (21 apps). Hourly tagged NOVEL; this is densify, not a + sibling first sighting. catalog ≠ endorsement. + +3. **[simota/tenbin](https://github.com/simota/tenbin)** + - DENSIFY neighbor skill (MIT; **2★**; HEAD `354aafa5d229`; + README SHA `90ac8803c13e`). Design-time lint / eval / thresholds. + Augustus does not absorb it. densify is not a sibling first + sighting. + +### HIGH (novel) + +4. **[kyegomez/open-jev](https://github.com/kyegomez/open-jev)** + - NEW HIGH measurement (Apache-2.0; **1★**; HEAD `fa57b06e23b8`; + README SHA `81139143b804`). Quote *theirs*: reconstruction ≠ replica. + unofficial research implementation with random weights. + kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ + Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev. Do **not** copy pip. + +5. **[jourdanlabs/assay-001](https://github.com/jourdanlabs/assay-001)** + - NEW HIGH measurement (license null; **0★**; HEAD `b7f71053a880`; + README SHA `eff9f28cfe3e`). Quote *theirs*: assay-001 split verdict. + CLINC150 ECE 0.0204 *theirs*. Banking77 ECE 0.0936 *theirs*. + 8,576 responses zero type errors *theirs*. Not Harbor. Do **not** + copy TypeSafe keys. + +6. **[brnyxx/jev-ra](https://github.com/brnyxx/jev-ra)** + - NEW HIGH measurement (MIT; **1★**; HEAD `ead323aa4fc2`; + README SHA `952bdd78cbc1`). Quote *theirs*: 3-5x / ~300 ms. + 8.50× Wikipedia *theirs*. systems comparison ≠ semantic + equivalence. Do **not** copy `uv` / CDP recipes. + +7. **[Promethe-us/awesome-jev](https://github.com/Promethe-us/awesome-jev)** + - NEW HIGH measurement (MIT; **15★**; HEAD `d1d7c0dcb05e`; + README SHA `cc682a221236`). Catalog, not a bench. + Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev. + catalog ≠ endorsement. + +8. **[ThePFMind/jev-mcp](https://github.com/ThePFMind/jev-mcp)** + - NEW HIGH measurement (license null; **0★**; HEAD `ef7733eb9822`; + README SHA `90593eb2cac3`). stdio, two tools: jev_evaluate, + jev_route. ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp. + serving substrate ≠ calibrated replica. Do **not** copy keys. + +### Remainder (short cards, same hour) + +- **[DragosTana/JEV-FC](https://github.com/DragosTana/JEV-FC)** : (empty description; name JEV-FC) (license null; **2★**; HEAD `a67ed576cf6c`). *theirs*. catalog ≠ endorsement. +- **[Nutlope/jev-fraud](https://github.com/Nutlope/jev-fraud)** : Use Jev with Kimi K3 for hard classification (license MIT; **2★**; HEAD `9b015f5782f5`). *theirs*. catalog ≠ endorsement. +- **[jeffloo886/jev-notion](https://github.com/jeffloo886/jev-notion)** : 🚀 Blazing-fast native macOS menu bar companion for Notion. Pure Swift & SwiftUI (<3MB), 8 languages. (license MIT; **2★**; HEAD `cbb7a215b2ad`). *theirs*. catalog ≠ endorsement. +- **[2456868764/jevguide](https://github.com/2456868764/jevguide)** : Curated Jev showcases from X, organized by category with media previews and direct source links. (license null; **1★**; HEAD `d68cc751a71f`). *theirs*. catalog ≠ endorsement. +- **[Solido/jev_dart](https://github.com/Solido/jev_dart)** : Typesafe Jev Api (license MIT; **1★**; HEAD `33b8ffe39c30`). *theirs*. catalog ≠ endorsement. +- **[andyhorn/jev](https://github.com/andyhorn/jev)** : (empty description; name jev) (license null; **1★**; HEAD `814d6a3e8214`). *theirs*. catalog ≠ endorsement. +- **[1npo/jev-gmail-labeler](https://github.com/1npo/jev-gmail-labeler)** : A tool that uses Jev to label my emails (license null; **0★**; HEAD `b4ec82ad88a1`). *theirs*. catalog ≠ endorsement. +- **[Aley3567/awsome-jev-sight](https://github.com/Aley3567/awsome-jev-sight)** : bright-sight 的独立优化副本：把一句中文指令变成受控的应用操作，每一步执行完都由代码回读验证 (license MIT; **0★**; HEAD `224d169a556e`). *theirs*. catalog ≠ endorsement. +- **[Carl-Lee91/Jev-VoC](https://github.com/Carl-Lee91/Jev-VoC)** : Jev 토이 프로젝트 (license null; **0★**; HEAD `dbe187b14b2b`). *theirs*. catalog ≠ endorsement. +- **[MartinPuli/f1](https://github.com/MartinPuli/f1)** : JEV Prix: five AI drivers, unknown procedural circuits, Formula-inspired racing, BYOK Jev and saved replays. (license MIT; **0★**; HEAD `83e0bcdf2eaf`). *theirs*. catalog ≠ endorsement. +- **[SeanPlusPlus/hellojev](https://github.com/SeanPlusPlus/hellojev)** : 👋 jev (license null; **0★**; HEAD `c3c419efafcf`). *theirs*. catalog ≠ endorsement. +- **[SuperInstance/jev-quilt](https://github.com/SuperInstance/jev-quilt)** : JEV for quilt as understood output: cellular-first decision substrate , typed cells, hook-and-drop deltas, bookkeeper WAL, last-mile projection decoupled (license null; **0★**; HEAD `a0f055593c9a`). *theirs*. catalog ≠ endorsement. +- **[asmirrr/DriftLab](https://github.com/asmirrr/DriftLab)** : Reproducible quantitative research CLI for testing momentum strategies and auditing research methodology with TypeSafe Jev. (license MIT; **0★**; HEAD `a6e750d17172`). *theirs*. catalog ≠ endorsement. +- **[bouncerguy/jevwrapper](https://github.com/bouncerguy/jevwrapper)** : English in. Typed JEV decisions out. Inspectable LLM-to-JEV middleware, a browser sandbox, and reusable examples. MIT licensed. (license MIT; **0★**; HEAD `17d0cd778052`). *theirs*. catalog ≠ endorsement. +- **[cristiancolon/jev-hft](https://github.com/cristiancolon/jev-hft)** : Research pipeline testing whether TypeSafe's Jev (via Vercel AI Gateway) can judge news and market data fast enough to matter. Bitcoin and US stocks, paper trading only. (license null; **0★**; HEAD `34a726317b4e`). *theirs*. catalog ≠ endorsement. +- **[digitalfoudnry-vb/JevBrowser](https://github.com/digitalfoudnry-vb/JevBrowser)** : JEV Browser (license MIT; **0★**; HEAD `6c9322302c3a`). *theirs*. catalog ≠ endorsement. +- **[dtduc-git/jevnav](https://github.com/dtduc-git/jevnav)** : Browser automation whose decisions you can replay, test and audit , Jev picks the element, traces replay offline in CI (license Apache-2.0; **0★**; HEAD `96f5438bea96`). *theirs*. catalog ≠ endorsement. +- **[gyozameronpanofficial/human-exe-jev](https://github.com/gyozameronpanofficial/human-exe-jev)** : HUMAN.exe , Jev-powered psychological checkpoint game (license null; **0★**; HEAD `dbb201dea114`). *theirs*. catalog ≠ endorsement. +- **[hawkyre/jevx](https://github.com/hawkyre/jevx)** : Find relevant conversations on X and score your drafts. Open-source Chrome and Firefox extension powered by Jev. (license MIT; **0★**; HEAD `50c311181da5`). *theirs*. catalog ≠ endorsement. +- **[kijung4290/maeum-on-attendance-care](https://github.com/kijung4290/maeum-on-attendance-care)** : 어르신 프로그램 출석 위험 모니터링 대시보드 · TypeSafe AI JEV 연동 (license null; **0★**; HEAD `0c845a7b57e1`). *theirs*. catalog ≠ endorsement. +- **[lafollett-labs/typesafe-jev-dojo](https://github.com/lafollett-labs/typesafe-jev-dojo)** : A live, graphical dojo for TypeSafe's Jev (System One) typed decision model , routing, a Tetris-playing agent, parallel swarms, and an honest Jev-vs-Claude gauntlet. (license MIT; **0★**; HEAD `0fc61808f691`). *theirs*. catalog ≠ endorsement. +- **[mateoromano-personal/contact-cleaner](https://github.com/mateoromano-personal/contact-cleaner)** : Sorts the thousands of addresses Google saved automatically into keep, review and remove. Next.js + Jev. (license null; **0★**; HEAD `be33eb9fdbe2`). *theirs*. catalog ≠ endorsement. +- **[nanami-0713/dsh-jev-decide](https://github.com/nanami-0713/dsh-jev-decide)** : DSH plugin: register TypeSafe Jev (System One decision model) as an agent tool , jev_decide returns calibrated probabilities (noul/choice/score) for routing/triage/guardrail judgments, no text generation. 把 TypeSafe Jev 决策模型注册为 DSH agent 工具 (license MIT; **0★**; HEAD `7209540b15c0`). *theirs*. catalog ≠ endorsement. +- **[peterrauscher/x-bookmark-sorter](https://github.com/peterrauscher/x-bookmark-sorter)** : Chrome extension (Manifest V3) that auto-sorts your X bookmarks into your own folders using TypeSafe's Jev decision model. (license MIT; **0★**; HEAD `7c27c711cf88`). *theirs*. catalog ≠ endorsement. +- **[sindunarasimhan/murmur](https://github.com/sindunarasimhan/murmur)** : Voice-first podcast listening with Jev and OpenAI. (license null; **0★**; HEAD `e089c25649a8`). *theirs*. catalog ≠ endorsement. +- **[treycausey/semantic-find](https://github.com/treycausey/semantic-find)** : Local evidence search with optional Jev semantic ranking (license MIT; **0★**; HEAD `f739609f9420`). *theirs*. catalog ≠ endorsement. +- **[ttlequals0/MinusPodJev](https://github.com/ttlequals0/MinusPodJev)** : MinusPod Jev Proxy (license MIT; **0★**; HEAD `ed5e8e0b7473`). *theirs*. catalog ≠ endorsement. +- **[yangzhou-chaofan/awesome-jev-prompt](https://github.com/yangzhou-chaofan/awesome-jev-prompt)** : latest top 100 showcases for jev (keep updating) from x / github / latest sources (license CC0-1.0; **0★**; HEAD `5071e882fa4a`). *theirs*. catalog ≠ endorsement. +- **[yasuhito/jev-computer-use](https://github.com/yasuhito/jev-computer-use)** : Safe computer-use automation guided by typed Jev decisions (license null; **0★**; HEAD `f5e99dea249a`). *theirs*. catalog ≠ endorsement. + +- **hf:samatv256/mini-Jev** : decision-model system-one. sha `6d142c6f062b` likes 1 apache-2.0. samatv256/mini-Jev ≠ r-ms/mini-jev. serving substrate ≠ calibrated replica. *theirs*. +- **hf:BlueAquilae/laya-pack-shaped** : safetensors region:us. sha `f5b89f979514` likes 0. serving substrate ≠ calibrated replica. *theirs*. +- **hf:akhilaaa3/openjev-v1-allmix-r512-merged** : gemma-4-12B-it merged r512. sha `558329a4e7ea`. hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo. serving substrate ≠ calibrated replica. *theirs*. +- **hf:cklxx/laya-browser** : laya system-1 browser-agent. sha `580ed2b8cc78`. serving substrate ≠ calibrated replica. Locate ≠ decide. *theirs*. +- **hf:frankyy03/laya-pt-es-nli** : laya pt/es NLI. sha `71459abd38f4`. serving substrate ≠ calibrated replica. *theirs*. + +Minesweeper cousins this hour (`comoc/jev-minesweeper`, +`EnesYilmazcode/JevMinesweeper`) are measurement notes only. +game success ≠ calibrated Noul. `namenu/pi-jev-effort` is a quota- +capped effort sibling of `TheoOliveira/pi-jev`, not a new species. +`Nutlope/jev-fraud` is Kimi K3 plus Jev for hard classification; +classifier ≠ authorizer. `jeffloo886/jev-notion` is a native macOS +companion. catalogs (`Promethe-us/awesome-jev`, +`2456868764/jevguide`, `yangzhou-chaofan/awesome-jev-prompt`) are +indexes. catalog ≠ endorsement. + +### Skips (thin / collision / name-match) + +- dabit3/jev-experiments already catalogued: densify, not a sibling. +- simota/tenbin already a neighbor skill: densify, not a sibling. +- kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev. +- Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev. +- ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp. +- namenu/pi-jev-effort ≠ TheoOliveira/pi-jev. +- samatv256/mini-Jev ≠ r-ms/mini-jev. +- comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper. +- hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo. +- Empty-README / 0★ playgrounds / name-match `Jev`: skip-thin. +- Archer rewrite: **promised_not_landed**. Hub archerhume/4rcherhume HTTP **401**. + +### Pulse (live REST this hour) + +Archer still promised_not_landed. Hub archerhume/4rcherhume HTTP **401**. +kev **1033★** (star-noise vs 1542 970; SHA move is the fold). +dabit3/jev-experiments **340★**. Promethe-us/awesome-jev **15★**. +tenbin **2★**. assay-001 **0★**. open-jev reconstruction **1★**. +This hour does not re-census SemIf / Laya likes / tracker; those numbers +stay §119 until a dedicated pulse. `invented_signal: false`. + +### Formal compose + anti-patterns + +Formal methods **compose** with scoring. A Noul is a SENSOR. A JSONL +file is labelled data, not Harbor. `--init_from` is a warm start, not +a replica of TypeSafe. A pytorch reconstruction with random weights +is an architecture sketch. A catalog is an index. A minesweeper win +is not a Noul. Treating 0.88 as Harbor, from-scratch as equal to +warm-start, kyegomez as razorback16, assay-001 as a global certificate, +8.50× as ECE, or Qwen3.5 as Archer is soundness theater. +from-scratch ≠ warm-start. JSONL labels ≠ Harbor. +reconstruction ≠ replica. catalog ≠ endorsement. +game success ≠ calibrated Noul. SHA move is not a replica. +Qwen3.5 ≠ Archer. + +### Adversarial review + testing hooks (Basit standing +order) + +Parent merge only after **CLEAN** adversarial review **AND** testing. +Hooks for the reviewer: + +- Uniqueness-gate: the consecutive `Hourly 1843 uniqueness lock:` + string must appear in every overlay listed below. Prior walls 0843 / + 0915 / jcr / 0922 / 0940 / 0947 / 1049 / 1143 / 1248 / 1340 / 1441 / + 1542 / 1643 / 1746 stay one substring each (do not mutate them; do not + reopen #23–#51). +- Namesake locks: Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ + yibie/awesome-jev; kyegomez/open-jev ≠ razorback16/openjev ≠ + TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; + ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; + namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; + samatv256/mini-Jev ≠ r-ms/mini-jev; + comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; + hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ + hf:akhilaaa3/openjev-r512-handoff-demo. +- Densify vs new: kev densifies §45. dabit3 / tenbin densify existing + mentions. Do not mint sibling first-sighting sections for those. + kyegomez / assay-001 / jev-ra / Promethe-us / ThePFMind are first + sightings this hour. +- Harbor-jevals: 0.794/0.832 / 0.812/0.837 / 0.33 vs 0.84 vs 0.83/0.88 + / ECE 0.0204 / ECE 0.0936 / 8.50× / ~300 ms are *theirs*, not Harbor. + Batch 5.8x / 4B 0.790/0.806 stay §126. 0.3.0 pin stays §127. +- Anti-patterns to refuse: TypeSafe drop-in; from-scratch as warm-start; + JSONL as Harbor; reconstruction as replica; catalog as endorsement; + minesweeper as Noul; 8.50× as ECE; Qwen3.5 as Archer; copying keys / + `npm` / `pip` / `uv` / `docker`. +- Overlay set: SKILL.md body (protocol fragments + class-table densify + + Hourly 1843), mental-models Apply 1843, composition-algebra items + 481–496, faq, mixed-architecture, validation, toolbox-mapping, + methods-catalog, formal-methods, formal-semi-formal, + applied-mappings, judgment-class, question-design, + agent-self-assessment, mappings, CHANGELOG, README, docs/ecosystem, + findings batch #111, refresh-log, sources.json, changelog-hourly.md, + revisit_fingerprints.json (kev SHA/README move; novel HIGH seeds). +- Offline check: `evaluate_decisions.py --self-test` (now includes + from-scratch ≠ warm-start / JSONL labels ≠ Harbor / reconstruction + ≠ replica) and `uniqueness_gate.py` (0843 + + 0915 + jcr / 0922 / 0940 / 0947 / 1049 / 1143 / 1248 / 1340 / 1441 / + 1542 / 1643 / 1746 / 1843). + No live Jev key. No wrappers. + +Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 diff --git a/research/refresh-log.md b/research/refresh-log.md index d30d6f9..558b57f 100644 --- a/research/refresh-log.md +++ b/research/refresh-log.md @@ -1,3 +1,20 @@ +## 2026-09-21 ~00:43 UTC / ~18:43 Boise - Hourly 1843 HIGH +- Fresh PR off latest `main` after merged #51 (hourly 1746 / `notes.md` §128 + / items 465–480 / batch #110) and merged #50 (hourly 1643 / `notes.md` + §127). Next free IDs: `notes.md` §129 / composition 481–496 / findings + batch #111. + **HARD RULE:** do not reopen or amend PR #23–#51. +- PRIMARY: jaredpalmer/kev own-data JSONL. --init_from warm-start LoRA/head PR #9. + from-scratch ≠ warm-start. JSONL labels ≠ Harbor. + Novel HIGH: kyegomez reconstruction / assay-001 split verdict / jev-ra *theirs* / catalogs. + SHA move is not a replica. +- Evidence: `research/archive/hourly/2026-09-21T00/` (novel HIGH + revisit HIGH). +- uniqueness_gate 0843+0915+jcr+0922+0940+0947+1049+1143+1248+1340+1441+1542+1643+1746+1843. + Evaluator: from-scratch ≠ warm-start / JSONL labels ≠ Harbor / reconstruction ≠ replica. + Quote *theirs*. No wrappers. `invented_signal: false`. +- Hourly 1843 uniqueness lock: jaredpalmer/kev densify HEAD bd058057ad0a README SHA 84b872488915; Fine-tuning on your own data; --data JSONL; --init_from warm-start LoRA/head PR #9; Kev-0.8B 4B 9B Qwen3.5 family; 4B new-source 0.794/0.832 *theirs*; 9B new-source 0.812/0.837 *theirs*; 0.33 vs 0.84 vs 0.83/0.88 *theirs*; from-scratch ≠ warm-start; JSONL labels ≠ Harbor; Kev-0.5B card Qwen3.5 family pointer; No Jev outputs were used for training; option order can change an answer; 8.2% ≥0.9 on wrong *theirs*; Kev-9B 7.5% ≥0.9 on wrong *theirs*; dabit3/jev-experiments densify 340★; simota/tenbin densify neighbor skill; Promethe-us/awesome-jev ≠ MrJev/awesome-jev ≠ yibie/awesome-jev; kyegomez/open-jev reconstruction ≠ replica; unofficial research implementation with random weights; kyegomez/open-jev ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev; jourdanlabs/assay-001 split verdict; CLINC150 ECE 0.0204 *theirs*; Banking77 ECE 0.0936 *theirs*; 8,576 responses zero type errors *theirs*; brnyxx/jev-ra 3-5x / ~300 ms *theirs*; 8.50× Wikipedia *theirs*; ThePFMind/jev-mcp ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper; game success ≠ calibrated Noul; Nutlope/jev-fraud Kimi K3; jeffloo886/jev-notion; hf:akhilaaa3/openjev-v1-allmix-r512-merged ≠ hf:akhilaaa3/openjev-r512-handoff-demo; serving substrate ≠ calibrated replica; catalog ≠ endorsement; SHA move is not a replica; wire-compat ≠ logit-equiv; Qwen3.5 ≠ Archer; Archer still promised_not_landed; Hub archerhume/4rcherhume HTTP 401; do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#31/#32/#33/#34/#35/#36/#37/#38/#39/#40/#41/#42/#43/#44/#45/#46/#47/#48/#49/#50/#51; notes.md §129 + + ## 2026-09-20 ~23:46 UTC / ~17:46 Boise - Hourly 1746 HIGH - Fresh PR off latest `main` after merged #50 (hourly 1643 / `notes.md` §127 diff --git a/research/revisit_fingerprints.json b/research/revisit_fingerprints.json index 8bf4fb8..e9f786b 100644 --- a/research/revisit_fingerprints.json +++ b/research/revisit_fingerprints.json @@ -20,7 +20,7 @@ "forks_count", "likes" ], - "note": "Last-look snapshots for already-catalogued sources. Hourly diffs these four fingerprints. Star-noise is not a fold. SHA move is not a replica. Seeded from merged notes (SemIf \u00a7117, NanoJev \u00a7115) plus hourly 1248 densify (openjev \u00a775, von \u00a749, verdict \u00a771, kev \u00a745, jeff \u00a760, TypeLLM \u00a7113) plus hourly 1340 densify (openjev sdk 0.7 + MLX 400, kev PLAN_Qwen35) plus hourly 1441 densify (openjev STE backends+Codiv dual serving) and 1441 first sightings plus hourly 1542 densify (TypeLLM README 3k\u219212k B, kev family new-source) and 1542 first sightings (pi-jev densify, jev-sentinel, tool-routers, leanest, jevals, MrJev, jev-firewall, jev-codex-approval, HF encoder/quanto). default_sha is the full HEAD. pushed_at is the GitHub push clock. description_hash is sha256[:12] of the GitHub/Space description, or null when notes do not quote it. README SHA is an optional extra, not a substitute for description_hash. plus hourly 1746 densify (TypeLLM truncated thinking + qwen35_small, kev 0.8B Qwen3.5 family, jev-pruner §53, jevassert §70, jev-packs §64) and 1746 first sightings (vexjoy, Canny, jev-engineering, five-lines, jev-table, kev-ane).", + "note": "Last-look snapshots for already-catalogued sources. Hourly diffs these four fingerprints. Star-noise is not a fold. SHA move is not a replica. Seeded from merged notes (SemIf §117, NanoJev §115) plus hourly 1248 densify (openjev §75, von §49, verdict §71, kev §45, jeff §60, TypeLLM §113) plus hourly 1340 densify (openjev sdk 0.7 + MLX 400, kev PLAN_Qwen35) plus hourly 1441 densify (openjev STE backends+Codiv dual serving) and 1441 first sightings plus hourly 1542 densify (TypeLLM README 3k→12k B, kev family new-source) and 1542 first sightings (pi-jev densify, jev-sentinel, tool-routers, leanest, jevals, MrJev, jev-firewall, jev-codex-approval, HF encoder/quanto). default_sha is the full HEAD. pushed_at is the GitHub push clock. description_hash is sha256[:12] of the GitHub/Space description, or null when notes do not quote it. README SHA is an optional extra, not a substitute for description_hash. plus hourly 1746 densify (TypeLLM truncated thinking + qwen35_small, kev 0.8B Qwen3.5 family, jev-pruner §53, jevassert §70, jev-packs §64) and 1746 first sightings (vexjoy, Canny, jev-engineering, five-lines, jev-table, kev-ane) plus hourly 1843 densify (kev own-data JSONL / --init_from warm-start, SHA 8465c4c4c294 → bd058057ad0a) and 1843 first sightings (assay-001, kyegomez reconstruction, jev-ra, catalogs).", "looks": [ { "id": "github:TheoLeeCJ/SemIf", @@ -85,14 +85,14 @@ { "id": "github:jaredpalmer/kev", "notes_section": "45", - "last_look": "2026-09-20T23:46Z", + "last_look": "2026-09-21T00:43Z", "fingerprints": { - "default_sha": "8465c4c4c2942e93f1458ff080624891febb6089", - "pushed_at": "2026-09-20T23:59:32Z", + "default_sha": "bd058057ad0aa9df3dd6d14e3542c95a5ce367b2", + "pushed_at": "2026-09-21T00:06:11Z", "description_hash": "47fc60928c90", "release_tag": "kev-family" }, - "readme_sha": "19664b9ae54672326c7aee14fef932e3eec7a58c" + "readme_sha": "84b8724889155d3fcc794abe96f8bf36700289c3" }, { "id": "github:logan-markewich/jeff", @@ -1326,6 +1326,210 @@ "release_tag": null }, "readme_sha": "90f887e3935f3e673223e7f39a901e46ee6f88cd" + }, + { + "id": "github:dabit3/jev-experiments", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "c469e5bfdc73eb3e1999bba2569e66b579a970fd", + "pushed_at": "2026-09-21T00:17:40Z", + "description_hash": null, + "release_tag": null + }, + "readme_sha": "ce01b8ab7f388c48c3b623675c09382381fde229" + }, + { + "id": "github:simota/tenbin", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "354aafa5d2299fd7f602bfa48db9bf72af073fec", + "pushed_at": "2026-09-21T00:15:16Z", + "description_hash": "b72a2a23ec3d", + "release_tag": null + }, + "readme_sha": "90ac8803c13e32ca3ce7ac41c9a8921924f33c26" + }, + { + "id": "github:Promethe-us/awesome-jev", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "d1d7c0dcb05efa88ecb190dac07868c1be1f4ba7", + "pushed_at": "2026-09-20T12:03:56Z", + "description_hash": null, + "release_tag": null + }, + "readme_sha": "cc682a221236ed8cac4d0eaa331fc5f4d980b05a" + }, + { + "id": "github:kyegomez/open-jev", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "fa57b06e23b8ca7444596744d4d118f7429749f8", + "pushed_at": "2026-09-21T00:24:14Z", + "description_hash": "6913694477a7", + "release_tag": null + }, + "readme_sha": "81139143b8040c8c6306f55384ebdb7d55a987ba" + }, + { + "id": "github:jourdanlabs/assay-001", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "b7f71053a88025a5af80af00801c5ce725c9547d", + "pushed_at": "2026-09-21T00:19:55Z", + "description_hash": "d602297bea69", + "release_tag": null + }, + "readme_sha": "eff9f28cfe3e79d3b0501fbc1c87e3c928d47d4f" + }, + { + "id": "github:brnyxx/jev-ra", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "ead323aa4fc2f906440a32cf9fd1494e6f7c00cf", + "pushed_at": "2026-09-21T00:19:33Z", + "description_hash": "0243f197153c", + "release_tag": null + }, + "readme_sha": "952bdd78cbc1d73628d2bec93b9da8d7a418a1f3" + }, + { + "id": "github:ThePFMind/jev-mcp", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "ef7733eb9822d3f705c2029827cb5470e01ca50d", + "pushed_at": "2026-09-21T00:16:02Z", + "description_hash": "b9bc5fd053eb", + "release_tag": null + }, + "readme_sha": "90593eb2cac31fc91b4365b5735c3d00c9e9e0e2" + }, + { + "id": "github:Nutlope/jev-fraud", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "9b015f5782f555424107f6b4bb9d67811df62b7d", + "pushed_at": "2026-09-20T03:58:09Z", + "description_hash": "bde9cd3c5ef8", + "release_tag": null + }, + "readme_sha": "480e2a626aa66bdad7a33f7147f316218476ea4b" + }, + { + "id": "github:jeffloo886/jev-notion", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "cbb7a215b2adf71250df69203c9c5af982ed53ea", + "pushed_at": "2026-09-20T07:11:21Z", + "description_hash": "9fa8121fc090", + "release_tag": null + }, + "readme_sha": "e3a8a6fcc81305e662f9c27934ef3c46a04921e5" + }, + { + "id": "github:namenu/pi-jev-effort", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "2e1d6f192c0bd878c70ff6ecf423340ec2a0b538", + "pushed_at": "2026-09-21T00:38:35Z", + "description_hash": "e6e4ba305098", + "release_tag": null + }, + "readme_sha": "1d9f79f764e1106c8adaafd959a7e9c04648f523" + }, + { + "id": "github:comoc/jev-minesweeper", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "5cbc361b1260cdfef6e9383639797bf483c064e0", + "pushed_at": "2026-09-20T12:50:17Z", + "description_hash": "4bedf0a9c9dd", + "release_tag": null + }, + "readme_sha": "bc8261842cfc14df7a8d5177fab2212421099d22" + }, + { + "id": "github:EnesYilmazcode/JevMinesweeper", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "c03f665b79ff5076916b389ec2467366e7aa51ef", + "pushed_at": "2026-09-21T00:43:17Z", + "description_hash": "7673715de975", + "release_tag": null + }, + "readme_sha": "274d4817cd26a354c62b8c07a11b92c7b9699562" + }, + { + "id": "hf:samatv256/mini-Jev", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "6d142c6f062bafaecbf8de30027f033ff3b4429b", + "pushed_at": "2026-09-21T00:23:41.000Z", + "description_hash": null, + "release_tag": null + }, + "readme_sha": null + }, + { + "id": "hf:akhilaaa3/openjev-v1-allmix-r512-merged", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "558329a4e7eac969dc4b81675b148a5f279f72da", + "pushed_at": "2026-09-21T00:11:37.000Z", + "description_hash": null, + "release_tag": null + }, + "readme_sha": null + }, + { + "id": "hf:cklxx/laya-browser", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "580ed2b8cc78a8b282cc1680b3571b60ca4fb1c0", + "pushed_at": "2026-09-21T00:18:34.000Z", + "description_hash": null, + "release_tag": null + }, + "readme_sha": null + }, + { + "id": "hf:frankyy03/laya-pt-es-nli", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "71459abd38f4e439b2f4e3a2dae87573ae40c77e", + "pushed_at": "2026-09-21T00:18:50.000Z", + "description_hash": null, + "release_tag": null + }, + "readme_sha": null + }, + { + "id": "hf:BlueAquilae/laya-pack-shaped", + "notes_section": "129", + "last_look": "2026-09-21T00:43Z", + "fingerprints": { + "default_sha": "f5b89f97951426763086f04fcdc814c8757bfa68", + "pushed_at": "2026-09-21T00:42:15.000Z", + "description_hash": null, + "release_tag": null + }, + "readme_sha": null } ] } diff --git a/research/sources.json b/research/sources.json index 00ac62c..b666790 100644 --- a/research/sources.json +++ b/research/sources.json @@ -1,6 +1,6 @@ { "refresh_cadence": "hourly", - "retrieved": "2026-09-20T23:46Z", + "retrieved": "2026-09-21T00:43Z", "sources": [ { "kind": "docs",