Augustus

Agent skill for placing TypeSafe Jev Choice/Score/Noul with classical decision methods, composition algebra, and a validation gate.

Augustus — named for Augustus De Morgan (1806–1871), mentor and professor of William Stanley Jevons — is the design-judgment skill for where typed probabilistic judgment belongs (the Jev-class of System One models), using mathematical, logical, and algorithmic mental models. It applies across AI, software, business, knowledge work, and life — not only SWE. TypeSafe Jev is the documented exemplar (Choice, Score, Noul), not the monopoly. Formal methods are one pillar. Exact work stays in code or policy; the model owns narrow judgment; never launder a Noul as a proof.

Companion, not replacement, to the official typesafe-ai skill. That skill owns Jev integration contracts; Augustus owns the design judgment: which pillar, family, and classical method map, what the objective implies for fail-open vs fail-closed, and what experiment would prove a design wrong. Not a TypeSafe-only how-to.

The skill

• .agents/skills/augustus/SKILL.md — working protocol + decision-design card

• .agents/skills/augustus/references/mental-models.md — cross-domain frames (EU, abstention, VOI, MCDA, SDT, search/control, Leveson, NATM/snap-fit/Norman); not SWE-only. Extractable-from-state boundary map (self-contained vs needs outside knowledge)

• .agents/skills/augustus/references/judgment-class.md — the class (Jev exemplar, not monopoly): open heads (Laya, kev, encoder DeBERTa, LoRA distill, domain specialist on independent gold), constrained-AR (TypeAR, pcdServer), announced decision-model (Watch), open multimodal RLCD (blackwood-rlcd; not Archer), Laya ONNX port, contract-compatible local /v1/systemone (stub until hf scorer; also kev pointer / von §49 Needle SAN snapshot ≠ this-pass 395M / n=78 — not replicas; jevify CUDA/PyTorch packed-logprob cousin — uncalibrated likelihoods ≠ Noul; jeff GLiFormer-400M encoder drop-in — not a Jev replica; sysone loopback gateway routes hosted + local, not a model; githubnext/localjev prompted JSON ≠ structured-read logits — ≠ kunchenguid/local-jev), GLiNER/GLiClass species (locate vs categorize vs local multi-head; GLiNER2.5 extractive compaction as a named job, not a new species; GLiNER code-graph indexer + escalate-S2, 10–50× unfilled; GLiNER2 observe→score-among-candidates computer-use as a different named job, not GLiNER2.5; typesafe-computer-use OCR+AX desktop hosted Jev of the same hole, never screenshot-to-frontier for the decision; gliner-native-runtime unofficial Swift/Core ML GLiNER 2.5-small on ANE — entity spans + confidence, not Choice/Score/Noul, not TypeSafe), Decision Graph Protocol (numerous-com/dgp; Jev-first assessor-neutral; app retains permissions/effects; guarded commit / receipt/next frame; not an official TypeSafe spec), kev family (Archer-arch fidelity; OOD 0.76–0.77 vs Jev 0.86; /v1/systemone drop-in is not a Noul; replica honesty), OpenJev local /v1/decide (not TypeSafe drop-in; distinct from hraness/sysone OpenJev runners), semif-serve SemIf /v1/systemone runoff (wire-compat ≠ replica), listwise vs decision objectives, vision scoring, when-to-use axes (including decision-model vs constrained LLM), agent-architecture portents

• .agents/skills/augustus/references/formal-methods.md — judgment vs proof ownership; Alloy Analyzer vs Apalache (finder ≠ BMC ≠ inductiveness); TLA+/Quint/P/NuSMV/PRISM/Event-B/mCRL2/KeYmaera; Dafny/JML/Frama-C/SPARK/ITP; DST trio (Antithesis hypervisor, Resonate HQ durable-async Lean+oracle+SDK, PufferLib env+seed); TOCTOU-of-Noul, soundness theater, AI×FM harms (Hillel, Cauli); NATM/snap-fit/Norman/ Leveson/Kent/Shirky

• .agents/skills/augustus/references/formal-semi-formal.md — one-screen alias of the FM pillar

• .agents/skills/augustus/references/mixed-architecture.md — default placement: judgment-class model + LLM + code; preference lint; provider (Jev default / other family with self-eval); dual-process S1 decide / S2 generate; component node; DOM-as-text + fan-out; shadow-mode compaction rollout; fail-open wake vs fail-closed merge-gate; Harbor on/off routing; hybrid local decide + remote fill; DONE ≠ verified success; evidence-preserving stdout prune (hard envelope then Noul; OpenCode host-port indiejoseph/opencode-jev-pruner, jev-zen / jev-1.13-free, zen-chat ≠ Noul); Decision Graph Protocol envelope (numerous-com/dgp; frame→assess→commit; app retains permissions/effects; Jev-first assessor-neutral; assessment batching; hard-gating DGP as safety theater); calibrated meaning-grep live tree (can1357/jegrep; path+range Nouls; no embeddings/index/daemon; ~$0.01–0.03 typical; agent --json); cost-derived YES/NO/UNSURE overlay (Kungie/gut; thresholds from costs not hard-coded; auto-batch same-object; overlay not species); typed-callback control flow (Illusion47586/judge; exactly one app-owned callback; explicit uncertain); variable-N option scoring (zwliJay/jev-forge; dynamic candidate bags; not a sixth species); open NAR replica late-catch (wfzyx/von; Needle snapshot ≠ 395M table; replica honesty); specialist S1 computer-use (Cua-S1 form-v0; plan ≠ execute; not TypeSafe Jev); judgment as a language primitive (hunch); decision-native RAG (retrieve wide → decide → evidence set); classify-first MCP (jev-sift); draft-gate heartbeat; living class-pattern atlas; public judgment wall; PR attention ≠ correctness; session-sticky first-prompt route; capability kernel (secrets never in agent; Jev SENSOR); typed control plane around DSPy; engine owns truth / Jev owns judgment; human-confirmed kill; decide→policy→LLM leftover cascade; wire-compat encoder backend; loopback gateway; closed-vote computer-use (no planner LLM); host-owned handlers × System One; active-learning triage (do not distill Jev as teacher); evidence-packet explorer; meaning-grep AND/OR/NOT; OMP prompt suppression (permission vs probability; operator owns the bar); judgment ≠ permission (skill-broker outline, not a recipe); constrained optimizer + S1 features (slo-router; never sole hot-path gate); effect-based shell gate (privilege ≠ verdict); attention filter / VOI (jev-lens; never blocks; never green unless sure); measurement owns endorsement (jev-packs evidence-gated + jevassert record/replay CI); Jev supplies evidence / code owns authority (actiongate-jev); ranking ≠ calibration (never hard-threshold raw p as frequency); hot-click CU (ego-jev; indexed table; S1 on click path); Jev judges relevance / code decides structure (jev-compactor); local rules first / never auto-train on own hides (x-reply-filter); control-plane combinators (not chat turns); skill VOI / abstention (skillranker hook fail-open); receipts not leaderboard (atlas); OOD / AUC ≠ ECE (sign flips by type); thinking-budget bake-off (frontier-100); turnstile evidence≠authority + replay; MLX one-pass replica economics (jevmlx; softmax ≠ Noul); never confidently wrong / TLA+ compose (jev-labs); no seal no advance / coverage ledger (seal; mint ≠ product brain); sureness bands (how-sure-is-jev; max_prob is generous); JevBench Harbor practice (calibration not in Main Score); CI typed gate before expensive review (ci-gatekeeper); Codex MCP host adapter (jev-in-codex); judgment as attention redirect (jev-preflight; not a merge blocker); compress-before-first-send (dizk/jev-lens; 79% fewer tokens); tools≠use / SessionStart over hoping (carryforward 0/4); observational memory (pi-om keep/kind verbatim); open-Jev class (openvons; JevPick; wire-compat ≠ replica); physical-world S1 (HA-Jev; not for locks); judgment outside the store (jevql CLI); landed-script trust / headless≠auto-approve (construct); digital-design combinators (jev-combinators rename + extended five); VOI cache admission (jevcache 0 FP/100); decision-ledger memoization (hyperspaceai/jevcache; fingerprint after redact; recall vs decide; Cache hit ≠ correctness; ≠ kushals256/jevcache); GEPA alignment loop (sutro-sh/jev-align; human labels only; score never auto-accepts; production capture flywheel; ≠ caiovicentino/jev-align); worth-your-attention VOI (ThinkyMiner/Winnow ≠ kevinpita/winnow); Jev WHETHER / Python HOW / LLM WHAT (hermes-jev-router); typed escalate/continue/abort baton (jev-handoff; gate never grants); Playwright executes, Jev chooses (browser-jev); OpenJev /v1/decide ≠ drop-in + SemIf runoff wire; conflict ≠ ignorance (named Choice escape); decision-as-memory flywheel (DGUI_HYPERMEM-JEV); record/replay CI (jevassert LANDED); measurement owns endorsement now has a runner (jev-packs 2,990-case matrix; calibration+cost first-class); failure-finding arena (jevarena ≠ jev-arena); BBQ stereotype/uncertainty/cost (not a bias cert); decider≠executor (jeffrey; pick ≠ fill); sentence-as-rule lint (mizchi/jev-lint is jevlint rename ≠ huntedman/JevLint); VOI hunk prune (prune-review ~20% cost target); whole-repo intent VERIFIED/VIOLATION/UNKNOWN; GLiNER2 System One spec ≠ replica; open replica substrates (grande / laya-jolt / JEV-CPU / local-jev measured not equivalent); githubnext/localjev prompted JSON ≠ structured-read logits; persist constraints across compaction (pi-heed); Harbor SGR-judge contract (jev-judge-bench; canaries ≠ quality; no headline yet; ≠ jevarena/jevbench); hand no-text steps (jev-use; Vercel drops confidence); Pi System-One control plane (pi-jev-control; GUI never force-click); never free-generates (jev-gpt tree of Choices); OpenRouter recipe atlas (jev-cookbook; samples not benches); personal-history feed (jevfeed; no social graph); competing NAR claim-audit (openJev-verdict-2.0; PR #1; ≠ OpenJev); empty compaction-proxy skip (IPECTER); 1-token logprob endpoint ≠ Noul (chakuho; coverage ≠ correctness); open replica engine (jevinf; argmax-parity ≠ ECE); 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; decision-native inbox (mailordinal); unofficial jev-cli not ready (≠ jevql); laya-multilingual English checkpoint confident-wrong OOD; schema-scorer peaked ranking ≠ calibration; productized System One HTTP (classifier.dev; label+confidence; batch ~1000); escalate-under-threshold (smart single-label <0.7; multi-label ignores); silent FALLBACK (granite 0.546 vs advertised 0.800; rh-guard owns the gate); systematic-review pointer (choxos/jev-reviewer ≠ egma-ai; two-pass Choice+Noul; human tick is the product); wire-compat ≠ logit-equiv (githubnext/localjev ≠ kunchenguid/local-jev; prompted JSON ≠ structured-read logits); Laya packaging ≠ new species (NandhaKishorM/laya; Router script-before-p; post-T ECE ≠ raw ECE; 0.85 still soft); external openjev census ≠ scored bake-off (@airesearch12; GLiNER2+routers class-boundary; incomplete vs watch; Harbor honesty watch); JevBench v1.2 geometric-mean I/C/S/K (cal ON rank; Jev 75.3 / SemIf 74.6 theirs; Luna I=96.8 rank #7; option-order 72→21; ×2/est. Harbor honesty; Laya absent gap; Qwen3.8 27B ≠ Archer); hourly 0842 already-folded recipe (wire≠logit · FALLBACK · packaging honesty · pointer-not-generator · leaderboard VOI; skip thin noise; hard-gate Noul as PR/quality = soundness theater); S1 never stalls / S2 one-use advisory (khordoo/jev-reflex-autonomy-lab delta; purple = consumed; Local controller ≠ githubnext/localjev; seed = geometry; 20% still soft; no pixels; S2 never grants); OCR+AX desktop observe→score→act (typesafe-computer-use; never screenshot-to-frontier for the decision; overlapping options = doubt; 155× theirs one screenshot; 0.4/0.5 still soft; ≠ jev-ultrafast ≠ cua-s1); ASR voice-browser observe→score→act (jev-voice-browser; never waveform-to-Jev; partial-speech VOI; spoken confirm ≠ auth; numbered overlay; 27/27 theirs fixtures; ≠ jev-voice-control ≠ nikolas-j ≠ OCR desktop); wrap-as-execution ALLOW/ASK/DENY (AgentGhost; wrap is the tool function; rules first; ASK throws; fail-closed; ≠ actiongate ≠ jev-use; rh-guard owns the gate cousin); JP genre atlas (@studio_yebisu; stars research-time ≠ eval; ≠ class census ≠ v1.2); external pedagogy / how-to-apply (@akshay_pachaar “Jev Clearly Explained”; LLM hammer; schema-safe ≠ correct; 200×/400× TypeSafe ceiling; shadow + questions-as-code; ≠ official docs ≠ Flavio); meaning-grep dedicated (jev-semgrep; proposition ≠ embedding; AND/OR/NOT after threshold; Semgrep.dev collision; not a gate; contrast-set refund); decision-validated UI (gram-render never authors text; jev2ui Jev decides / Gemini writes; valid ≠ good); decision-as-assert (jevtest ambiguous band; 0.85 still soft; record/replay); hybrid S1 (anima3 closed verb menu + hard safety; Qwen logprob; jeff confidently flat; do not invent Laya); pointer search (JevFind path then window); Harbor bake-off trio (jev-frontier-bench ≠ frontier-100; GLiClass product bakeoff; four engines / majority floor / calibration ≠ discrimination); authorship named escape (not evidence); non-SWE (ha-switchboard HA remains execution ≠ HA-Jev; n8n Low Confidence); compaction delta (fast-jev-compaction-pi ≠ pi-jev-compact ≠ pi-jev-compaction); full-distribution optimizer (jevloop UCB1+CEM; no LLM in the loop; mock default); deferred class (laya-vision SmolVLM score untrained ≠ blackwood ≠ Archer; Cerebellum-2B /v1/decide ≠ TypeSafe / wire-compat vs agent-routing; laya-grounded not drop-in / Platt not temperature); open LoRA replica (GestaltLabs/Jeff-1 ≠ logan-markewich/jeff; acc/ECE tradeoff n=9730 theirs); Jev-first bounded agent (stanley-code; empty ≠ approve; human promote); NL memory → beam-search FS (findme ≠ JevFind); price workers not the conversation (jevsubrouter; fail-open; counts ≠ dollars); Typed if (feelings .feels() default 0.5 is Noul-0.5-never-rounded; ≠ hunch ≠ Probably); Shadow then honor (apa-agent-harness ≠ AntonioCoppe/jev-harness; grok-bot-jev skill honor / A/B proxies ≠ tokens); Human every action (Essentiel-Jev never authority); Atom then sense (enzo-mcp independently falsifiable claims; ≠ jev-sift); File by Choice (pigeonhole OTHER skip; HF playground static ≠ classifier.dev); Question preflight (jev-reliability Nothing about accuracy; clduab11/jev-test bars ≠ scores; jev-rag-benchmark “Jev wins” is not an assumption; dairui1/jev-lab ≠ BrendanH18/jev-lab); Inbox read-only vs write (jevmail gmail.readonly; mailjay archive/trash after review; ≠ mailordinal)

• .agents/skills/augustus/references/applied-mappings.md — context sieve, exact-text keep/drop (extractive / pointer-not-generator; char-offset compaction; observed a11y/DOM controls; Bash stdout prune; verbatim session ledger / carryforward 0/4 tools≠use; classify-first MCP / jev-sift; Stagehand extract pick-and-copy; jevcumber meaning-as-spec; closed-vote JevOnly; host-owned waymode; jev-compactor framework-agnostic compact+gate 73% product-arm; dizk/jev-lens pre-send views; pi-om observational keep/kind), env triage (OpenSmoke + latch merge-gate; ci-gatekeeper pre-review typed gate; jev-preflight Stop-hook attention redirect, not a merge blocker), moderation/ranking (decision-native RAG evidence set; living class-pattern atlas; meaning-search without embeddings / jevgrep; meaning-grep jev-semgrep; evidence-packet jevex; measured RAG rerank vs generative rerank; sift ~$0.00003/post; ThinkyMiner/Winnow worth-your-attention VOI ≠ kevinpita/winnow), skill routing (route ≠ memory; session-sticky first-prompt lock; OMP/pi fail-open jev_route; OMP prompt suppression / omp-greenlight; skill-broker outline — Jev never grants access; slo-router constrained optimizer + S1 features; skillranker VOI / hook fail-open; jev-in-codex Codex MCP adapter; pi-jev-skill-bench Harbor roster-size harness; pi-jev-skill-suggestion strip-roster), capability kernel / human-confirmed gate (interlock vs toolgate; port-cleanup; permission vs probability; spoken confirm ≠ auth / jev-voice-browser; wrap-as-execution ALLOW/ASK/DENY / AgentGhost — wrap is execution; ASK throws; fail-closed; rh-guard owns the gate cousin; construct-auto-classifier privilege ≠ verdict + landed-script / headless≠auto-approve; actiongate-jev — Jev supplies evidence, code owns authority; turnstile — policy first, replay; seal — no seal no advance / coverage ledger; jev-labs — never confidently wrong; jev-handoff typed baton — gate never grants), decide→policy→LLM leftover cascade (jav-email-cascade), closed-vote computer-use (applied-mappings §9; ego-jev hot-click cousin; browser-jev Playwright executes Jev chooses; jeffrey decider≠executor; jev-use hand no-text steps; jev-gpt never free-generates; typesafe-computer-use OCR+AX desktop cousin; jev-voice-browser ASR voice-browser cousin), VOI hunk prune / whole-repo intent (prune-review / jev-intent-review), persist constraints (pi-heed; Jev never writes policy), Pi control plane (pi-jev-control), personal-history ranking (jevfeed; ≠ Winnow), OpenRouter recipe atlas (jev-cookbook), empty compaction-proxy skip (IPECTER), Pi verbatim summarizer replacement (pi-jev-compact ≠ pi-jev-compaction), decision-native inbox (mailordinal), commit pre-review (commitjev), jevex n=16 files-to-read, productized classification API (classifier.dev; spam/inbox/feedback), systematic-review pointer (choxos/jev-reviewer ≠ egma-ai; two-pass; human check never overwritten), prompted-JSON local /v1/systemone (githubnext/localjev ≠ kunchenguid/local-jev; wire-compat ≠ logit-equiv), Laya packaging (NandhaKishorM/laya Router over Hub ckpts; not a TypeSafe drop-in), external openjev census (@airesearch12 / Benchmark Heaven; tweet ≠ v1.1 ≠ live ranks), JevBench v1.2 scored board (geo-mean I/C/S/K; cal ON; 534 decisions; ≠ v1.1 87.6), hourly 0842 apply-the-five (already §73–§78; skip thin; soundness-theater PR gate), continuous-control S1/S2 (khordoo delta §80; escalate without stall; Local ≠ localjev), OCR+AX desktop CU (typesafe-computer-use §81; exclusive actions; split kind/item/site; writer/decider; 155× theirs one screenshot), ASR voice-browser CU (jev-voice-browser §82; partial-speech VOI; pointer spans; spoken confirm ≠ auth; 27/27 theirs fixtures), wrap-as-execution ALLOW/ASK/DENY (AgentGhost §83; wrap is execution; ASK throws; fail-closed; rh-guard owns the gate cousin), JP genre atlas (@studio_yebisu §84; stars research-time; not verified evals; ≠ class census ≠ v1.2), external pedagogy (@akshay_pachaar §85; schema-safe ≠ correct; 200×/400× TypeSafe ceiling; questions-as-code), meaning-grep dedicated (jev-semgrep §86; proposition ≠ embedding; boolean composition after threshold; Semgrep.dev collision; not a gate), hourly 1047 + deferred 0945 (notes.md §87; decision-validated UI; decision-as-assert; hybrid S1; pointer search; Harbor trio; authorship; ha-switchboard ≠ HA-Jev; n8n; compaction-pi namesake lock; jevloop; laya-vision / Cerebellum / laya-grounded), queued SIGNALs + jevsubrouter (notes.md §88; GestaltLabs/Jeff-1 ≠ logan-markewich/jeff; stanley empty ≠ approve; findme beam ≠ JevFind; jevsubrouter prices workers), hourly 1144 (notes.md §89; feelings typed if; apa-agent-harness ≠ jev-harness; grok-bot-jev skill honor; Essentiel-Jev never authority; enzo-mcp ≠ jev-sift; pigeonhole OTHER skip; playground static; jev-reliability Nothing about accuracy; clduab11/jev-test ≠ jevtest; jev-rag-benchmark “Jev wins” is not an assumption; dairui1/jev-lab ≠ BrendanH18; jevmail readonly / mailjay writes after review), hourly 1241 (notes.md §90; ZHUBoer/ego-jev reserved __none__ / runWorkflow completed ≠ success ≠ jiangkoumo; 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), hourly 1347 (notes.md §91; judgekit YAML classify/score/route/verify; typed-judge-kit verdict-in-code; alsoleg89/decide packing VOI / 0.8 ≠ 80% accuracy ≠ jev-sift; Jev-Calibration Platt ECE 0.117→0.052; jev-calibration-arena never acts ≠ jev-arena; ctmx/openrouter-jev-mcp Decision-as-Plugin; cyrusasco/typesafe-mcp noul deadband 0.35–0.65; FrancoisChastel/jev-code ≠ npm jev-code; claudecode-jev-marketplace fail-open not hot path; pedroknigge/mcp_jev packs not ask_jev; 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), hourly 1441 (notes.md §92; 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; rh-guard owns gates; Soft Noul ≠ hard safety), SIGNAL jevcache/jev-align (notes.md §93; 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; memoize typed decisions; VOI of cache hit; GEPA + System One; Soft Noul ≠ hard safety), SIGNAL enzyme/JA ModernBERT/Gemma (notes.md §94; 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; Soft Noul ≠ hard safety), hourly 1541 (notes.md §95; 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; rh-guard owns gates; Soft Noul ≠ hard safety), hourly 1639 (notes.md §96; 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; Soft Noul ≠ hard safety), SIGNAL gliner-native-runtime (notes.md §97; 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), hourly 1740 (notes.md §98; 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; Soft Noul ≠ hard safety), hourly 2145 (notes.md §102; 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; 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; Soft Noul ≠ hard safety), hourly 2246 (notes.md §103; independent System One evidence catalog; 19 reviewed records; 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; Soft Noul ≠ hard safety), hourly 2041 (notes.md §101; 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; Soft Noul ≠ hard safety), hourly 1943 (notes.md §100; 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; probably is a language not an overlay; Soft Noul ≠ hard safety), hourly 1843 (notes.md §99; 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; NAR local drop-in; open replica economics / latency vs closed Jev; 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; overlays not species; Soft Noul ≠ hard safety)

• .agents/skills/augustus/references/faq.md — "just classification", stack replacement, Jev vs open head vs encoder vs LoRA vs constrained AR vs kev vs blackwood, wait-for-Archer, missing-other confident-wrong, soft project rules vs linter (Abide), extractive/pointer-not-generator, compaction summarize vs pointer, encoder vs Jev compaction, fail-closed keep_full, shadow-mode rollout, fail-open vs fail-closed wake vs CI gate, observe→score→act backend-agnostic, hybrid local decide + remote fill, DONE ≠ verified success, stdout prune vs session compaction, Cua-S1 vs TypeSafe Jev, plan ≠ execute / dry-run, local drop-in vs stub scorer, route ≠ memory, when-it-holds / extractable-from-state, decision-model vs constrained LLM, dual-process S1/S2, S2 never flies / Local controller ≠ localjev / purple = consumed, combinatorial grid ≠ extractive, GLiNER vs GLiClass vs CLIP, LLM-as-judge, in-engine vs CLI store, hard envelope (bitrate / planner), not-another-how-to, uncalibrated local likelihoods ≠ Noul, decision-native RAG, classify-first MCP, living applied-mappings atlas / class patterns, draft-gate silence ≠ safer, robotics text-state vs pixels, Stagehand extract pick-and-copy / fast-path not replacement, public judgment wall / six parallel questions, meaning-search without embeddings, attention≠correctness PR review, skills→oxlint not a hard gate, session-sticky fail-closed routing, measured RAG rerank vs generative rerank, cascade sign-flip / calibration theater, Precision PDF honest negative, type-safe ≠ correct / Jev is SENSOR not policy, Ax/DSPy knobs vs typed control plane, native vs verbalized confidence, engine owns truth / Jev owns judgment, train specialist vs few-shot hosted, Noul 0.5 cannot-tell never rounded, calibration ≠ sortable, local /v1/systemone ≠ Jev (GLiFormer / gateway), do not distill Jev as teacher of record, PCD O(1) ≠ calibrated Noul, closed-vote CU vs Stagehand pick, OMP/pi fail-open vs pi-jev-approver, permission vs probability / operator owns the safety bar, judgment ≠ permission / Jev never grants access, eval integrity / instrument not score, Jev not sole hot-path gate / constrained optimizer + S1 features, privilege ≠ verdict / effect contracts, attention filter not permission, measurement owns endorsement / evidence-gated packs, Jev supplies evidence / code owns authority, ranking ≠ calibration / never hard-threshold raw p as frequency, Jev done ≠ browser success, never auto-train on the model's own hides, pointer compact ≠ LLM summarize, combinators not a new model, AUC ≠ ECE / sign by type, thinking-budget bake-off, local MLX one-pass ≠ Noul, never confidently wrong / TLA+ compose, no seal no advance, sureness vs max_prob, JevBench calibration not in Main Score, CI typed gate before expensive review, Codex MCP adapter, two jev-lens products, Stop-hook not merge blocker, tools≠use, openvons not TypeSafe, Noul not for locks/heaters, two Winnow products, OpenJev /v1/decide not drop-in, SemIf runoff ≠ replica, combinators rename + extended five, Noul conflict≠ignorance, jevcache fail-open, typed baton never grants, jevassert record/replay CI, jevarena≠jev-arena, BBQ not a bias cert, jeffrey pick≠fill, jev-lint is jevlint rename ≠ JevLint, local-jev not equivalent, constraints survive compaction (pi-heed), jev-judge-bench≠jevarena≠jevbench (no quality headline yet), jev-use≠ultrafast / Vercel drops confidence, pi-jev-control GUI never force-click, jev-gpt never generates, cookbook samples not benches, jevfeed no social graph, openJev-verdict claims ≠ OpenJev / not endorsement, 1-token logprob ≠ Noul / coverage ≠ correctness, jevinf replica ≠ TypeSafe, elixir-sdk ≠ dannote/jev, jevex n=16 rename, commitjev middle band, hermes-plugin-jev is Agnes, pi-jev-compact ≠ pi-jev-compaction, IPECTER runway empty, mailordinal inbox, jev-cli not ready ≠ jevql, laya-multilingual confident-wrong OOD, schema-scorer peaked ranking, HF 401 / GitHub 404 Hub-only, classifier.dev productized HTTP / escalate-under-threshold / silent FALLBACK / vs_jev tracked JSON, choxos/jev-reviewer ≠ egma-ai / two-pass Choice+Noul / not-found / human tick is the product, githubnext/localjev ≠ kunchenguid/local-jev / wire-compat ≠ logit-equiv / prompted JSON ≠ structured read / 1200-req bake-off caveats, NandhaKishorM/laya packaging ≠ Hub-only / Router script-before-p / post-T ECE ≠ raw / Banking77 token-budget / 0.85 still soft / vs-Jev unpublished-here, @airesearch12 census ≠ jevbench v1.1 / GLiNER2+routers class-boundary / incomplete vs Laya-localjev-kev / likes ephemeral, JevBench v1.2 geo-mean I/C/S/K / cal ON rank / 75.3 vs 87.6 not a drop / Luna I=97 rank #7 / option-order 72→21 / ×2 latency assumption / Laya absent gap / Qwen3.8 27B ≠ Archer, hourly 0842 already-folded / apply-the-five / skip thin noise / hard-gate Noul as PR gate is soundness theater, screenshot-to-Jev for CU / 155× Harbor score (typesafe-computer-use: no, and no), waveform-to-Jev / 27/27 Harbor score (jev-voice-browser: no, and no; spoken confirm ≠ auth), AgentGhost sidecar / ASK skip (no, and no; wrap is execution; ASK throws; fail-closed), JP genre atlas bake-off / live ★ (studio_yebisu: no, and no; stars research-time; not verified evals), Akshay how-to / 200× Harbor (no, and no; TypeSafe ceiling; schema-safe ≠ correct), jev-semgrep Semgrep.dev / embedding tricks / a gate (no, no, and no; proposition ≠ embedding; boolean after threshold; ranking fail-open), Jev writes UI text / valid GramSpec is good (no, and no), jevtest 0.85 as product proof (no), anima3 default jeff / invent Laya (no), Jev 72.5% class ceiling (no), GLiClass architecture duel (no), skip majority floor (no), authorship as evidence (no), ha-switchboard replaces HA-Jev (no), n8n official TypeSafe (no), compaction-pi = compact (no), jevloop mock quality (no), laya-vision is Archer (no), Cerebellum typesafe-sdk drop-in / endorse 94.92% (no), laya-grounded drop-in / temperature (no), Jeff-1 is logan-markewich/jeff / better ECE than Jev (no, and no; n=9730 theirs; set reused), empty stanley findings as approval / auto-promote (no; notChecked; Soft Noul ≠ hard safety), findme is JevFind / beam proves the file (no), swap conversation model / quote jevsubrouter dollars (no; counts ≠ dollars), .feels() a new language / default 0.5 a bool if (no, and no), apa-agent-harness is jev-harness / copy @aipersona (no, and no), grok-bot-jev 13.0× as token savings / skill forces the bot (no, and no), Essentiel Jev send mail / 0.75 calibrated (no, and no), enzo-mcp is jev-sift / skip UNKNOWN (no, and no), pigeonhole OTHER as a move / 0.6 Harbor τ (no, and no), HF playground live Jev / classifier.dev (no, and no), jev-reliability measures accuracy (no), clduab11/jev-test is jevtest / D already passed (no, and no), jev-rag-benchmark showed Jev wins (no), dairui1/jev-lab is BrendanH18 / re-card jev-desktop (no, and no), jevmail is mailordinal / mailjay read-only (no, and no), ZHUBoer/ego-jev is jiangkoumo / completed is success (no, and no), jsort scores are frequencies / Choice for scale (no, and no), groundedness-judge-bench showed Jev wins quality (no), jev_playground 83% is quality (no), copy jev-latest on Zen (no), techstack classifier generates a stack (no), s1_ruby is hunch / is? is a proof (no, and no), judgement is jevql / confidence is winner p (no, and no), Rust community SDK is official / a new species (no, and no), tpellet/hunch is carldaws/hunch (no), file-search 15 matches is recall (no), linkmap referee is gold / Jev sees S2 prose (no, and no), jev-mail is jevmail / tidy OTHER moves / close pinned tabs (no), ORIGIN LLM decides / continue without Jev (no, and no), crawlers raw bug_likely as a gate (no), jevbrain is TypeSafe Jev / 95.2% class number (no, and no), judgekit is JudgeBench / 97.7% class ceiling (no, and no), openrouter-jev-mcp is TypeSafe first-party (no), Gemini self-confidence as a threshold (no), decide 0.8 is 80% accuracy (no), calibration arena acts / is jev-arena (no, and no), raw Choice 50–95% is accuracy (no), FrancoisChastel/jev-code is stanley npm (no), marketplace Jev on the hot path / fail-closed missing key (no, and no), mcp_jev invents ask_jev (no), jevtypesafeai.com is TypeSafe (no), ts_safety is a Noul (no), hermes-switchyard is hermes-jev-router / Agnes / loads skills (no), nanoprune is hosted Jev / 0 hallucination (no, and no), jev-browser-agent is ZHUBoer / omp DONE is proof (no, and no), hari007sh/jev is dannote/jev (no), system-one-skills is a judge / 8,026 class ceiling (no, and no), typed-gate 0.51 is a yes (no), pi-jev-gate is fail-closed (no), Foq 100%/ECE 0.2% class ceiling / 25 ms Harbor (no, and no), rev measured replica / quote 32.4 (no), robfrase/jev running local / collapse into dannote (no, and no), typesafe_agent_gates 27/27 Harbor / Noul as hard deny (no), safe-sh pre-exec allow/block (no), pastepilot act without Confirm / clipboard watch (no, and no), Jev-Reranker 0.1667 live Jev / confidence scales value (no, and no), sessionwise required sieve / fail-closed down (no, and no), jev-search scores as truth (no), 400 ms Salesforce SLA (no), scheduler plugin places Pods / demo agreement as accuracy (no, and no), Android screenshot-to-frontier / collapse droidjev into ultrafast (no, and no), jevcu closed-vote / Jev writes the plan (no, and no), ha-conversation-jev is HA-Jev / copy OAuth client_id (no, and no), dsh-jev widen tools / live Jev default (no, and no), classification-benchmark $0.46 measured (no), luna-pagerduty 1.000 production paging (no), meldecision new Laya species (no), laya-doom sees pixels / is Archer (no, and no), quote laya-api README (no), akpsahan vs-Jev as new measure / Qwen3.8-27B Archer (no, and no), Jev owns chess truth (no), jev-drive is AV (no), Jev authors story-arc (no), auto-file HS6 (no), SC2 API Victory as UI win (no), awesome-jev-use-cases likes as eval (no), typesafe-go official (no), enzyme when asked as PreToolUse deny / catalysts as summaries (no, and no), enzyme hosted bootstrap as silent TypeSafe (no), unofficial JA ModernBERT is TypeSafe / skip format_version / collapse Argos1111/jev_local into us/jev-local (no), LFM default is JA ModernBERT (no), Nemotron_Jev is a calibrated Jev replacement (no), djev-dev is djev-spark / ships pixels as TypeSafe CU (no, and no), Laya essay vs-Jev is a new bake-off / Khmer 0.952 conf is competence (no, and no), OpenCode jev-pruner is tamaratran or fast-jev-opencode (no, and no), zen-chat is a Noul / paste 24/24 onto OpenCode (no, and no), fail-closed the OpenCode turn / keepThreshold 0.5 as proof (no, and no), jev-webagent-bench scores / JSON Schema boolean as a proof / JevSwiftSDK official (no, no, and no), gliner-native-runtime is TypeSafe Jev / Fastino official / Choice/Score/Noul (no, no, and no), collapse it into gliner25-compaction / gliner2-ultrafast / Eran-BA / JevSwiftSDK / jevmlx (no), README 0.99 as Harbor / hard-gate 0.1 as NER quality / invent ANE ms (no, no, and no), file it as keep/drop / position 4 Selector (no, and no), numerous-com/dgp is official TypeSafe / assessment p grants an effect / a receipt proves the decision was right (no, no, and no), hard-gate DGP as a safety proof / collapse ThreadDesk mocks into live Jev (no, and no), can1357/jegrep is Bentlybro/jevgrep / uehaj/jev-semgrep / paste 79% (no, no, and no), hard-gate 0.4/0.2 as concept absent / paste $0.01–0.03 as a class ceiling (no, and no), kev OOD 0.76 is Jev / isolation 4e-6 proves identity / /v1/systemone wire is a Noul / Archer landed (no, no, no, and no), Jev IS the if-statement / playground recordings (southpolesteve/probably: no, and no), re-dump jev-align / 133★ identity (no; SHA unchanged), retrieve by resemblance / 17/18 Harbor (no, and no), hard-gate 0 of 157 / questions as evidence (no, and no), jev-lint is a second product / fail CI on a shipped warning (no, and no), boolean @ 0.5 is a proof (no), ANE 4.98 ms beats Jev / claim 10× (no, and no), snapjudge softmax is a Noul / is localjev (no, and no), JevPi 62 tests are quality / is jevpilot (no, and no)

• .agents/skills/augustus/references/mappings.md — classical-method mappings with boundaries, counterexamples, acceptance tests (including Hypothesis cards §6–§19 — promote only with a test that ran)

• .agents/skills/augustus/references/validation.md — design gate, eval recipes, Jev-for-skills (routing, self-monitoring, testing, modularity, frontmatter), and Eval & hill-climb (jevals hygiene + Harbor taskset; open-jev-laya-bench as ECE/NLL/Brier bake-off exemplar; DMB as Harbor-style frozen protocol vs constrained LLMs; jevals-data as CC-BY-4.0 recompute-from-logs feedstock; Abide replay as Harbor-adjacent soft-rule measurement; solari-reflex Harbor-style computer-use; gliner2-ultrafast encoder-backend cousin (DONE ≠ success; demo is not a bake-off); Cua-S1 specialist form source-only (metric names, no checkpoint scores; not TypeSafe Jev); Stagehand extract pick-and-copy 37/75 no-LLM ~0.5s vs 4.37s (their card; pick ≠ replacement; draft #2951–#2955); jevgrep 79% top-5 vs BM25 / grep on stripped repos; Jev-RAG one-run vs Spark rerank (full-context Spark still faster); jev-oxlint Phoenix answer-key; native-probability calibration arena (jev-arena live Brier 0.0059 / ECE 0.0620 theirs); typed control-plane bake-off shape (jev-dspy-control-plane; offline stubs ≠ quality); jev-testbench collab arms; ARC-AGI Direct Jev as combinatorial-≠-extractive negative; jev-gateway-bench Harbor on/off routing one-run signal; jev-pruner Harbor needle/noise + Terminal-Bench integration pilot, not a full bench; jev-baselines-eval pre-registered AMBIGUOUS + cascade sign-flip; explore-typesafe-ai synthetic FHIR Harbor-shaped, not clinically validated; databricks-jev-pdf-lab honest negative; Domain-jev-maker specialist vs few-shot (KL/r/McNemar); jav-email-cascade compare arms; jev-orderby-bench ORDER BY gates (calibration ≠ sortable); jeff GLiFormer cost/accuracy; system-one-benchmark Jev vs MLX PCD vs AR JSON n=50 (Brier 0.1096 vs 0.3884); jevex 1/8→6/8 SWE finish n=8; jev-semgrep 0.94/0.98 (10×51 theirs; dedicated §86; 51★ ephemeral); omp-greenlight 1,013/10 default 40.9% / 0 of 94; dinostomp jev-as-if ECE 0.062 theirs / FINDINGS 189; slo-router p95 77.93→490.38 same routes; construct-auto-classifier Jev 0 dangerous / 975; INSTRUCT_JEV 119-row instruct seed; jev-packs nine verified packs on pinned jev-1.13 + jevassert LANDED (2,990-case matrix; Jev/Sonnet 5 accuracy tie, Jev better calibrated 7/9, ~250× cheaper; sms-spam 0.953/0.040); BBQ 58,492 / 97.28% / $0.3429 theirs; jev-lint (jevlint rename) 13/15 1.00/1.00; grande JGLUE 0.614/0.853; local-jev done 30%/shape 57%; pi-heed 98.5%/0 false block; does-jev-confidence 8,000 judgments AUC ~0.91 / stated ~75% vs human ~10% / ~96% ECE removed; ego-jev n=3 medians ~2× vs per-step LLM; jev-compactor 64.5%/ 366ms/0 invented paths vs Sonnet summary, one session; jev-frontier-100 Jev 77.0% vs Qwen3.5 4B/2048 96.7% (exploratory); jev-ood-calibration 900 tickets ECE 0.107 = 4.4× floor / sign flips by type; jev-labs 1,080 golden 0 wrong under chaos (escalate; not a proof of zero); jevbench v1.1 Jev 1.13.0 Main 87.6 (calibration not scored); how-sure-is-jev Choice confidence = max_prob; ci-gatekeeper 504–629 ms own-repo; dizk/jev-lens 79% / 500 trajectories; jev-compactor product-arm 73%/350ms/4 of 4; carryforward 0/4 recall; openvons JevPick 3.2–4.8×; HA-Jev 17★ not for locks; jev-preflight fail-open 8 axes; jevcache 0 FP/100; zeroshot-vs-bert +0.05–+0.13 / DiD; ThinkyMiner/Winnow 80%/90%; OpenJev 45/60; semif-serve 1164 vs 178 ms; typed-evaluation-collapse Noul vs named Choice; jev-judge-bench SLA-150 contract / canaries ≠ quality / no headline yet; jev-use 220 ms p50 / 12/12 / Vercel 0.4 theirs; jev-cookbook 425/$0.015 samples not benches; openJev-verdict-2.0 77.10%/0.0636/ 0.0144 theirs unverified + PR #1 claim-audit; chakuho GUI 336 27B 95%/92% vs Jev 89%/82% theirs; jevinf 2.57×/2.27× 100% argmax; jevex n=16 160s→69s / $8.74→$3.13; commitjev 0 false on 5 clean theirs; laya-multilingual MASSIVE 0.366/0.387 vs 0.227/0.733; schema-scorer v2 Choice 0.841; HF 401 this pass; classifier.dev 400/650 ms; F1 0.887 / 230 ms vs 0.799; AG News 87.7%; granite 0.546 vs advertised 0.800 theirs; NandhaKishorM/laya T4 32.8 ms / post-T ECE 0.081 vs Jev 0.246; Banking77 0.425 vs 0.870; 0.766 fine-tune theirs; @airesearch12 census tweet (engagement ephemeral; not a scored bake-off); JevBench v1.2 Jev 75.3 / SemIf 74.6 theirs; Luna I=96.8 rank #7; cal ON; ≠ v1.1 87.6; hourly 0842 recipe already §73–§78 / skip thin / soundness-theater PR gate; khordoo/jev-reflex-autonomy-lab Local-vs-Live A/B, not a scored bake-off; AgentGhost wrap-as-execution, not a quality bench (MIT 2★; ASK throws); @studio_yebisu JP genre atlas tweet (engagement ephemeral; stars research-time; not verified evals); @akshay_pachaar “Jev Clearly Explained” (engagement ephemeral; 200×/400× TypeSafe ceiling; schema-safe ≠ correct); uehaj/jev-semgrep meaning-grep dedicated (51★ ephemeral; 0.94/0.98 theirs 10×51; Semgrep.dev collision; not a gate); hourly 1047 + deferred 0945 (gram-render / jevtest / jev2ui / anima3 / JevFind / jev-frontier-bench 72.5% ECE 0.161 vs Fable 84% ECE 0.064 theirs / jev-gliclass-bench 78/40/49 / job-posting-triage floor 0.947 / authorship / ha-switchboard ≠ HA-Jev / n8n-nodes-jev / fast-jev-compaction-pi ~50× theirs / jevloop mock / laya-vision 75.2% ECE cal 0.034 / Cerebellum competing NAR not endorsement / laya-grounded phishing regress); queued SIGNALs + jevsubrouter (notes.md §88; GestaltLabs/Jeff-1 acc 0.8183 ECE 0.0807 vs Jev 0.8283/0.0932 n=9730 theirs ≠ logan-markewich/jeff / stanley empty ≠ approve / findme ≠ JevFind / jevsubrouter counts ≠ dollars / Soft Noul ≠ hard safety); hourly 1144 (notes.md §89; jev-reliability noul-gate 0.0%/12.5%/3.6% theirs Nothing about accuracy; clduab11/jev-test Nothing runs yet; jev-rag-benchmark “Jev wins” is not an assumption; dairui1/jev-lab urgent 91% vs Haiku 79% theirs synthetic; grok-bot-jev A/B proxies not tokens); hourly 1241 (notes.md §90; groundedness-judge-bench native vs schema-guided Jev 0.6667 vs GLM 0.7661 theirs; jev_playground 0 promotions / routing-backtest 0.0447%; jsort CommonLit r=0.824 / ρ=0.841 theirs; tpellet/hunch NL2Bash 36/120 theirs; jev-file-search scores not calibrated accuracy; jev-linkmap v1→v3 45%→65% theirs; ORIGIN pause-if-no-Jev; jevbrain AUTO_ACT is not a Noul); hourly 1347 (notes.md §91; judgekit 97.7% n=130 theirs; typed-judge-kit 7.0× / MIN_LABELS=20; decide 500 issues $0.0203 theirs / 0.8 ≠ 80% accuracy; Jev-Calibration Platt ECE 0.117→0.052; jev-calibration-arena never acts; FrancoisChastel/jev-code 1★ ≠ npm jev-code; hermes-switchyard ≠ hermes-jev-router ≠ hermes-plugin-jev; nanoprune 2.8MB ECE 2.58%; omp-jev-web 5.9s/0 vs 47.6s/27 theirs; typed-gate 0 wrong/0 omit/117 review theirs); hourly 1441 (notes.md §92; Foq 25 ms P50 / 2.2 GB / ECE 0.2% / 100% 150-case theirs; rev sample 32.4 not a bench; robfrase/jev 85–114 ms p50 theirs planning memo; typesafe_agent_gates 27/27 / 31/31 theirs; Jev-Reranker r@1 0.1667 offline not live Jev; jev-search 24p $0.01/7.4s theirs; 400ms Salesforce WebMCP demo timestamps ≠ Harbor; droidjev find ~0.6s/iter theirs; jevcu smoke 12s / 324–380 ms theirs; ha-conversation-jev FAST_MIN 0.80 theirs; classification-benchmark specified not run; luna-pagerduty p≥0.50 1.000/$0.062 theirs n=3000; meldecision tokenize 6.9 / infer 312.4 ms theirs; jevchess ~300 ms / game <1¢ theirs; SC2 UI-verified Liberation Day 3:44 / Outlaws 27:57 theirs ≠ API Victory; akpsahan/laya ≠ Archer / do not re-paste Nandha vs-Jev), SIGNAL jevcache/jev-align (notes.md §93; Cache hit ≠ correctness; score never auto-accepts; hyperspaceai/jevcache ≠ kushals256/jevcache; sutro-sh/jev-align ≠ caiovicentino/jev-align), SIGNAL enzyme/JA ModernBERT/Gemma (notes.md §94; guidance ≠ hook; unofficial ≠ TypeSafe; Nemotron ≠ TypeSafe Jev; not a calibrated replacement; JGLUE JNLI 92.62% / JComQA 92.40% theirs; Gemma ~0.2s theirs; Laya essay numbers theirs), hourly 1541 (notes.md §95; orchestrator finish 0.95 / test 0.8 theirs sensors; replacement 20-case acc 0.90→1.00 F1 0.9028→1.0000 p50 675.66→253.13 ms cost 1.52× theirs failed cost gate; quarry 5 s fail-open / p<0.5 drop / top 3; OpenRoboto apple-to-plate 113 cycles $0.018825 vs Astra $5.933624 theirs one seed-0 not a rate; one-dollar-tahoe ~74 demo no ASR/FPR; llama-jev 80 ms cold / 40 ms cache theirs softmax ≠ Noul), hourly 1639 (notes.md §96; OpenCode jev-pruner unit tests ≠ Harbor; do not copy tamaratran 24/24 / 83%; jev-webagent-bench empty stub), SIGNAL gliner-native-runtime (notes.md §97; README 0.99 fixture; no Harbor; default threshold 0.1 still soft), hourly 1740 (notes.md §98; DGP 106 tests ≠ Harbor; mock resolver ≠ Jev; jegrep no published Harbor do not copy 79%; OpenRouter/TypeSafe auto-failover is silent FALLBACK; kev family OOD 0.76–0.77 vs Jev 0.86 theirs; replica honesty; Score confidence is a stand-in), hourly 1843 (notes.md §99; von n=78 / 93.0% theirs ≠ Harbor; do not merge Needle 52.6%; llm-vs-jev RESULTS generated from summary.json ≠ Harbor; nothing wins outright; competing NAR claims / replica honesty; jev-forge 0.579/0.637 theirs not Harbor), hourly 1943 (notes.md §100; pointer mode 17/18 19/20 theirs ≠ Harbor; 0 of 157 false invalidations tuned on same set; 189/189 FP16 checkpoint parity ≠ task accuracy; 62 tests wiring not quality; 10× not achieved; softmax over allowed tokens ≠ Noul), hourly 2041 (notes.md §101; 1824 independent Nouls / callback 25.0% theirs ≠ Harbor; Jev 0.587 / PrismNLI 0.725 / Laya 0.587 theirs ≠ JevBench; contamination caveat; 123,805 requests / ECE 0.075 / 47% unanswerable @≥0.95 theirs ≠ jevals; historical-v0 zero retained cases; on small inputs lenses lose money; quote both $0.000004 and $0.000006), hourly 2145 (notes.md §102; 0.75 vs 0.81 theirs ≠ measured separation; 62 cargo PASS ≠ Laya parity; unpaired 0.577 vs 0.727 ≠ JevBench; AUROC 0.769 vs acc 50.1; +0.199 / +26,744 / 0 of 2,816 theirs ≠ Harbor; maze 1.0000 n=141 research artefact; exit 0 ≠ claim truth; MagaBitmex/jev-4b-distill model not found), hourly 2246 (notes.md §103; TokenTrim no-Jev matched hybrid 62.4% theirs ≠ bake-off; 21 tasks · 134 items · 208 questions constructed ≠ production; Hev 80.00%/ECE 0.020 theirs development-only; mini-jev 93.25% self-authored ≠ family-disjoint; residual-head 9,222-param decreased 73/96→67/96; ChatJev-style soundness theater; timeout = censoring; ranking ≠ calibration / 0.5 still soft))

• .agents/skills/augustus/references/boundary-audit.md — existing-system insertion: fit test, opportunity map, smallest boundary, red flags

• .agents/skills/augustus/scripts/evaluate_decisions.py — offline evaluator for selective binary decisions (Brier, reliability, threshold/cost sweep)

Plus research/ — the living evidence archive behind the skill, refreshed hourly (see research/README.md).

Install

Claude Code (plugin marketplace, mirrors the official TypeSafe layout):

claude plugin marketplace add 24601/Augustus
claude plugin install augustus@augustus

Any skills-compatible agent (Amp, Codex, Cursor, …):

npx skills add 24601/Augustus --skill augustus

ChatGPT: skills are not a native ChatGPT primitive — paste .agents/skills/augustus/SKILL.md plus the references/ files into a GPT's instructions or a Project's knowledge and it will follow the protocol.

Amp: repo-local .agents/skills/ are discovered automatically.

GitHub topics

jev typesafe typesafe-ai system-one system-one-models structured-output calibrated-confidence ai-agents agent-skills decision-systems reranking beam-search claude-code python llm decision-theory semantic-search agent-workflows mixed-architecture tool-routing skill-routing semantic-lint classification gliclass listwise-ranking vision-scoring open-weights formal-methods model-checking deterministic-simulation decision-theory value-of-information signal-detection mcda calibration alloy apalache pufferlib stamp-stpa

Versioning

See CHANGELOG.md and releases. Current: 0.3.0, written against typesafe-ai/skills v0.5.7 (65a39f3). Re-read live TypeSafe docs before treating that pin as current API behavior.

License

MIT — see LICENSE.