Fold HIGH NiazMorshed2007/jcr: capability-tree lookup, does not execute

.agents/skills/augustus/references/agent-self-assessment.md

@@ -931,3 +931,8 @@ User-provided 0806 uniqueness lock: people who compare Jev against GPT-5.6 has n
 
 Hourly 0843 uniqueness lock: A hunch is a probability with a policy attached; { enter: 0.8, exit: 0.6 } is hysteresis; replay a policy change without inference; Decision models are providers, not the product; huncho ≠ Kungie/gut ≠ carldaws/hunch ≠ tpellet/hunch; pretrained Qwen2.5 base ECE 0.030 (0.5B) / 0.040 (7B); instruct 0.302 / 0.269; 70.9% → 70.0% mean conf 74.1% → 96.7%; temperature scaling still matches it in-distribution; No Jev API was called; Qwen2.5 ≠ Archer; Qwen/Qwen3.8-27B ≠ Archer; 学習済みモデル v0.1 は準備中です; bool AUROC 0.523; 先頭だと0件、末尾だと250件; 温度を渡さない場合、確率は較正されていません; このリポジトリには Jev を呼ぶコードが存在しません; g0runmezadam/what-is-jev IS tunahansahin897/what-is-jev (same GitHub id 1378007307); 947 repos scored; A 273 · B 302 · C 372; LLM rubric ≠ benches; Data as of 2026-09-20; HEAD 895b9498; README SHA 3ae98c56; 13 focused checks and one mutually exclusive outcome; Probabilities are advisory, not calibrated guarantees; omni-/ask-jev ≠ pedroknigge/mcp_jev; pd.cut bins by equal width while jeval bins by quantile; ECE 0.113 and ECE 0.076; jeval drift is not implemented yet; rlaope/jeval ≠ dayhaysoos/jevals; calibration does not compose; ECE has exactly zero statistical power to detect the failure mode that kills trajectories; 25–60× headline withdrawn; P(all-correct): 0.0071 vs 0.0001; TCE / AMS; Qwen 3.8 sparring ≠ Archer; Deferred Crispification; light_cutoff_applied_to_combination 0; recorded run, kinematic animation; BANKING77 Accuracy BERT-Base 93.02 Jev 79.90; Analyse jev calibration (NLL, ECE) backlog; BERT figures are published supervised references, not zero-shot; 档位措辞效应 分数极差中位 0.50、最大 1.32; 修好后对照组是 0.01; 不是 benchmark; 概率没做 calibration; ~1,430 API calls, about $0.15; xiaohuaxi/jev-study ≠ wjdjdakf17/jev-study ≠ baekenough/jev-study; AND: product (independence assumed and recorded in the trace); chat model's stated confidence is not calibrated; circuit-vl-4b ≠ Archer; Bring your own API key; vamsikrishna2421/jev-usecases ≠ whyashthakker/awesome-jev-use-cases; catalog ≠ endorsement; SemIf 2237★ (+30 vs §111 2207); jevlike 1049★ (+6 vs 1043); TypeLLM/TypeLLM 16★; AnotiaWang 98★; yibie/awesome-jev 520★ (+14 vs 506); Laya likes 861 (was 822); tracker likes 66 (+2 vs 64) lastModified UNCHANGED; Blackwood likes 2 gated manual; 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; notes.md §114
 User-provided 0915 uniqueness lock: TianyuCodings/NanoJev unified-games-v1 densify; A 0.6B parallel decision model: states and questions in, complete probability distributions out. Zero output-token decoding.; One model, four games; ViZDoom Basic 128/128 vs Jev 56/128; Predict Position 27/128 vs Jev 11/128; Maze 225 attempts vs Jev 2738; Snake 30 food / 256 steps; held-out Maze 4/10 Snake 8/8 Basic 128/128 Predict 27/128; Untuned Qwen3-0.6B baseline; 18,760 questions per variant; 16,333 ViZDoom; 896 Predict Position expert episodes; hard_lr1e5; mix weights 1/3, 1/3, 1/6, 1/6; Hub C-Tianyu/NanoJev revision unified-games-v1 likes 58; dataset C-Tianyu/NanoJev-Data likes 5; HEAD 618cea6d906d54e128360786d12f703fff2b1245; 1289★ / 158 forks / size 64035; README SHA 4190093c64ee75b26e9726daa3b00cbcf6d3157a; MIT; caijinchun/nanojev-arena ≠ liao96312/jev-arena-nanojev ≠ zwliJay/jev-forge ≠ NanoJev; not TypeSafe Jev; open replica / specialist gameplay S1; soft scores ≠ hard gates; Game success ≠ calibrated Noul; local type boolean ≠ TypeSafe noul; invented_signal false; do not reopen or amend PR #31/#32/#33/#35; notes.md §115
+
+
+**0920 jcr:** pre-action lookup is not a done-check and not a permission gate. Ambiguity / no-match / depth-limit are explicit abstention paths. NiazMorshed2007/jcr. `notes.md` §116.
+
+User-provided 0920 jcr uniqueness lock: NiazMorshed2007/jcr MIT; site https://jcr.niazmorshed.dev; topics ai-agents,jev,mcp; **4★**; HEAD `138b3832`; README SHA `2a49dbc1`; LICENSE SHA `46231303`; size **14850**; Jev Capability Resolver; one tool to find documented deterministic commands in a nested capability tree; returns context; **does not execute**; skills = workflow+judgment; capabilities = individual operations; format independent of Jev; proposed open standard exploration; classify (Jev) → optional OpenAI decompose compound → beam search geometric mean of routing probs; keep up to 3 paths ≥60% of best (JCR_BAND_RATIO 0.6); ambiguity / no-match / depth-limit explicit; soft scores ≠ hard gates; 0.6 band is application policy; routing ≠ permission; docs ≠ authority to run; sol-vs-opus5-20 *theirs*: 20 scenarios × 4 variants = 80 runs; lookup+explain only, no execution; Claude Opus 5: agent input 108,585→15,819 (−85%), cost $0.3700→$0.1222 (−67%), wall 105.5s→77.7s; Codex GPT-5.6-Sol: 61,952→47,669 (−23%), $0.1377→$0.1151 (−16%), wall 25.3s→62.4s (Sol slower with JCR in 19/20); One Sol outlier 372.6s / 193 Jev calls; n=1 per cell; Not Harbor task-execution; Claude/Codex harnesses; compare mode; 50 scenarios bundled; 11 groups, 960 nodes, 11,360 items; 16 routing rounds per step; 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

.agents/skills/augustus/references/applied-mappings.md

@@ -2421,6 +2421,11 @@ lastModified UNCHANGED `2026-09-20T04:29:16.000Z`;
 do not reopen or amend PR #23/#24/#25/#26/#27/#28/#29/#30/#32/#33.
 Do not copy keys / `npx` / `pip` / `uv` / `.env` / `TYPESAFE_API_KEY`. Soft Noul ≠ hard safety.
 
+
+### 0920 jcr capability resolver (`notes.md` §116)
+
+Place JCR as retrieve-wide → decide → evidence-set on a nested capability tree. Skills stay workflow+judgment; capabilities are individual operations. The resolver returns documentation; the harness owns execution, credentials, and checks. Soft scores ≠ hard gates. routing ≠ permission. docs ≠ authority to run. NiazMorshed2007/jcr ≠ skill-broker ≠ skillranker ≠ jev-sift ≠ jev-lens ≠ jevusher ≠ jev_select_capability.
+
 Hourly 0743 uniqueness lock: 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/#33; notes.md §113
 
 **Hourly 0646 HIGH (`notes.md` §111).** Measurement PRIMARY + datasets/Spaces + applied/skills/economics:
@@ -2497,3 +2502,6 @@ not TypeSafe Jev; open replica / specialist gameplay S1. Game success ≠ calibr
 caijinchun/nanojev-arena ≠ liao96312/jev-arena-nanojev ≠ zwliJay/jev-forge ≠ NanoJev.
 Do not copy `pip` / `snapshot_download` / `serve_decisions`.
 User-provided 0915 uniqueness lock: TianyuCodings/NanoJev unified-games-v1 densify; A 0.6B parallel decision model: states and questions in, complete probability distributions out. Zero output-token decoding.; One model, four games; ViZDoom Basic 128/128 vs Jev 56/128; Predict Position 27/128 vs Jev 11/128; Maze 225 attempts vs Jev 2738; Snake 30 food / 256 steps; held-out Maze 4/10 Snake 8/8 Basic 128/128 Predict 27/128; Untuned Qwen3-0.6B baseline; 18,760 questions per variant; 16,333 ViZDoom; 896 Predict Position expert episodes; hard_lr1e5; mix weights 1/3, 1/3, 1/6, 1/6; Hub C-Tianyu/NanoJev revision unified-games-v1 likes 58; dataset C-Tianyu/NanoJev-Data likes 5; HEAD 618cea6d906d54e128360786d12f703fff2b1245; 1289★ / 158 forks / size 64035; README SHA 4190093c64ee75b26e9726daa3b00cbcf6d3157a; MIT; caijinchun/nanojev-arena ≠ liao96312/jev-arena-nanojev ≠ zwliJay/jev-forge ≠ NanoJev; not TypeSafe Jev; open replica / specialist gameplay S1; soft scores ≠ hard gates; Game success ≠ calibrated Noul; local type boolean ≠ TypeSafe noul; invented_signal false; do not reopen or amend PR #31/#32/#33/#35; notes.md §115
+
+
+User-provided 0920 jcr uniqueness lock: NiazMorshed2007/jcr MIT; site https://jcr.niazmorshed.dev; topics ai-agents,jev,mcp; **4★**; HEAD `138b3832`; README SHA `2a49dbc1`; LICENSE SHA `46231303`; size **14850**; Jev Capability Resolver; one tool to find documented deterministic commands in a nested capability tree; returns context; **does not execute**; skills = workflow+judgment; capabilities = individual operations; format independent of Jev; proposed open standard exploration; classify (Jev) → optional OpenAI decompose compound → beam search geometric mean of routing probs; keep up to 3 paths ≥60% of best (JCR_BAND_RATIO 0.6); ambiguity / no-match / depth-limit explicit; soft scores ≠ hard gates; 0.6 band is application policy; routing ≠ permission; docs ≠ authority to run; sol-vs-opus5-20 *theirs*: 20 scenarios × 4 variants = 80 runs; lookup+explain only, no execution; Claude Opus 5: agent input 108,585→15,819 (−85%), cost $0.3700→$0.1222 (−67%), wall 105.5s→77.7s; Codex GPT-5.6-Sol: 61,952→47,669 (−23%), $0.1377→$0.1151 (−16%), wall 25.3s→62.4s (Sol slower with JCR in 19/20); One Sol outlier 372.6s / 193 Jev calls; n=1 per cell; Not Harbor task-execution; Claude/Codex harnesses; compare mode; 50 scenarios bundled; 11 groups, 960 nodes, 11,360 items; 16 routing rounds per step; 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