[24601]

Docs-only fold of hourly System One / categorization watch 1241 (Boise ~12:41 MDT) into Augustus how-to-apply, mental models, mixed architecture, toolbelt, and Harbor/jevals practice. Not a thin Jev skill dump. Not a hit list.

PR #3 / #4 / #5 already merged. This is a fresh PR off main. Never reopen merged PRs. Do not re-fold 1144 / §89.

Archer Hume 27B drop is still WATCH (HF empty; tracker likes 49 lastModified 2026-09-19T18:37:18Z still promised). Laya yes. Blackwood ABSENT. Census user-provided, not re-derived: SemIf 1846 (+17), jevlike 962 (+3), TypeAR-AI/TypeAR 10 (+1), Awesomejev flat 561/27007.

Non-negotiable: exact work stays in code or policy; the model owns narrow judgment; never launder a soft Noul as a proof. Formal methods compose with categorization/scoring; a Noul is a SENSOR. Hard-gating AUTO_ACT, treating ranking logits as frequencies, pasting 95.2% / 83% plumbing / 36/120 NL2Bash as class ceilings, or letting Jev send/delete/close pinned tabs is soundness theater. Soft Noul ≠ hard safety. No wrapper, no SDK/plugin how-to, no copied TYPESAFE_API_KEY / ZEN_API_KEY / AI_GATEWAY / uv / cargo / wrangler / chrome unpacked / OAuth client ids / shop URLs / .env. No invented metrics. TypeSafe Jev is the exemplar, not the monopoly. Backend-agnostic categorization/scoring/decision class (Jev-like speed/econ).

Lane is Augustus only (mental models / architecture / Harbor-jevals / toolbelt). 0★ HIGHs still get real cards because they were named HIGH.

This update — 18:41 UTC / ~12:41 Boise 2026-09-19 (research/notes.md §90)

Fifteen HIGH repos plus a five-repo product cluster.

1. Observe→score→act (namesake lock)

ZHUBoer/ego-jev (MIT; 0★; README SHA fc4537ac) — Ego Lite observe/act; Jev choose over compact page state. Reserved __none__. choose does no act and applies no universal cutoff. runWorkflow completed ≠ success. Exact work local. ≠ jiangkoumo/ego-jev.

2. Decision-as-ranking

keltokhy/jsort (MIT; 1★; README SHA a7c1f705) — pairwise Jev → Bradley-Terry. Scores are relative logits, not frequencies. Noul not Choice for scale. CommonLit r=0.824 / ρ=0.841 theirs. Fed hawkish ρ=+0.47 (91) / +0.38 (87) theirs.

3. Native vs schema-guided Harbor

slavadubrov/groundedness-judge-bench (LICENSE MIT / SPDX NOASSERTION; 0★; README SHA 5cc070d8) — RAGTruth QA. Native Noul vs schema-guided LLM JSON. Fastest/cheapest ≠ quality. Jev Macro-F1 0.6667 vs GLM 0.7661 theirs. implicit_true included in yes (world-true still unsupported-by-passages). ≠ jev-judge-bench ≠ jevarena ≠ jevbench.

4. 0 promotions / authored vs real

JYeswak/jev_playground (MIT; 0★; README SHA 96483039) — 0 promotions on purpose. A suite that passes a random judge is plumbing (254/305=83%). routing-backtest 0.0447%. ≠ HF playground.

5. Offload + classifier-not-generator

yuyang2230/jev-agent-skill (MIT; 0★; README SHA 158ca045) — leftover writer vs cheap decide via OpenCode Zen. Pin jev-1.13-free. ≠ GodsBoy/jev-agent-skill-router.

swap-mitra/jev-techstack-classifier (license null; 0★; README SHA cd32bde4) — ranks options listed in stack_config.json only. Classifier not generator.

6. Collapse late

innocentdiaz/s1_ruby (MIT; 1★; README SHA 6250f97b) — S1 measures; ? collapses; undecided? abstains. Code asks; code decides. ≠ carldaws/hunch ≠ feelings.

7. Unofficial toolbelt

2389-research/judgement (license null; 0★; README SHA 23c2dcda) — Go CLI. confidence ≠ winner p. Pin vs 24h alias. ≠ jevql.

community-ports/typesafeai-sdk-rust-community (MIT; 0★; README SHA bda3ad23) — community Rust SDK. Not a new species. crates.io typesafeai-sdk-community 0.3.1.

8. Pointer shell

tpellet/hunch (MIT; 0★; README SHA b25d81ac) — pick/why/is/run over stdin/PATH/man. Exit 3 abstains. never-execute list. ≠ carldaws/hunch.

9. Preview-first VOI / rubric rewrite

emilwagman/jev-file-search (MIT; 0★; README SHA b2a236bb) — scores not calibrated accuracy. Recall unmeasured. ≠ JevFind.

stas4000/jev-linkmap (LICENSE MIT / SPDX NOASSERTION; 0★; README SHA 271c1cd2) — TF-IDF candidates; Jev never sees S2 prose.

10. Life fail-open covers

muhammedilyasy/jev-mail (metadata only; ≠ jevmail), MANISH007700/tidy (none-of-folders stay), MANISH007700/tab-bouncer (pinned/audio/current never closed), stefw/lkclean (Show fail-open), jaibhasin/jev-yt-time-saver (Show anyway).

11. S1 decide / S2 plan

JacquesGariepy/ORIGIN-CIVILIZATION (LICENSE MIT / SPDX NOASSERTION; 1★; README SHA 33e6ba07) — Jev every voluntary action; LLMs plan never decide. pause-if-no-Jev. validResponse sums-to-1. ≠ Essentiel-Jev.

12. Seed/expand/judge/verify + local daemon ≠ Jev

russfranky/jev-crawlers (MIT; 0★; README SHA 2d0b552e) — risk bands never raw boolean. Verify grounding, not exec.

Synxneuos/jevbrain (license null; 9★; README SHA 55fe0abf) — local n-gram/anchor overlap. AUTO_ACT is not a Noul. ≠ TypeSafe Jev ≠ classifier.dev.

Files

SKILL.md mapping index unique fragments; mental-models.md Apply 1241; mixed-architecture.md fail table + gallery; faq.md; judgment-class.md; validation.md; applied-mappings.md; mappings.md; toolbox-mapping.md; composition-algebra.md items 49–61; question-design.md; methods-catalog.md; formal-methods.md; agent-self-assessment.md; CHANGELOG / README / docs/ecosystem.md; research/notes.md §90; refresh-log.md; archive/findings.md batch #73; sources.json (524 / 521 unique, retrieved 2026-09-19T18:41Z).

Checks

• evaluate_decisions.py --self-test ok
• SKILL.md YAML frontmatter parses
• sources.json 524 sources / 521 unique (3 pre-existing dups unchanged)
• git diff --check clean
• Adversarial review on 5916012 FAIL then fixed on 4ae5085: jsort Fed ρ +0.47 (91) / +0.38 (87); implicit_true included in yes; ZHUBoer off jiangkoumo hot-click / closed-vote rows; count 15 HIGH + 5-app cluster.
• Did not invent a classifier.dev URL for jevbrain. Did not reopen merged Fold hourly 1047 HIGH + queued SIGNALs + deferred 0945 into Augustus #3/Fold hourly 1144 HIGH into Augustus mental models and jevals #4/Restore 1144 README quotes (UNKNOWN backticks, Jev-wins colon) #5.

  

[cursor]

Adversarial review — FAIL on 5916012, fixed on 4ae5085 (not merge-blockers now)

Read-only first on gh pr view 7 / gh pr diff 7 vs origin/main (after merged #4 / #5 / #6). Actual fold is fifteen 1241 HIGH receipts + a five-app product cluster into research/notes.md §90 and skill docs (mental models / mixed architecture / Harbor-jevals) — not a thin Jev skill dump. Spot-checked cited numbers against pinned READMEs / WHITEPAPER / EVAL.md / VERDICT.md at the SHAs in research/notes.md §90. evaluate_decisions.py --self-test → self-test ok. YAML frontmatter parses. Did not merge.

Checklist

#	Gate	Result
1	Invented metrics / unattributed numbers	FAIL → fixed. jsort CommonLit r=0.824 / ρ=0.841 / SMOG 0.661/0.647 / Choice 93% vs Noul 28% match README SHA a7c1f705. Fed hawkish was invented: PR said ρ +0.46 (95 statements) / +0.37; pinned README is +0.47 (91) vs that-day DFEDTARU and +0.38 (87) vs 180-day. Now research/notes.md:14311-14313, references/faq.md:2316-2317. Other theirs numbers checked: groundedness 0.6667 vs 0.7661 / 900/150 / 3,600 / $0.3611 (5cc070d8); playground 254/305=83% / 0/20 vs 3/20 / AUC 1.000 vs 0.750 / 0.0447% / VERDICT 17 ideas 0 promotions / demo-1 zero Jev (96483039 + VERDICT.md); tpellet 36/120 / 66 abstain / 15/20 / 739 ms / 6 ms (b25d81ac); file-search 1,070 / 15 / 7.576s / $0.0209 (b2a236bb); linkmap 566 / 8,460 / 5.95s / $0.27 / 45%→65% / 287/679 (271c1cd2); ORIGIN 961/0 planner / 192+38 (33e6ba07); crawlers n=12 / $0.000035 / dogfood 14/5/0 / U7 unvalidated (2d0b552e + EVAL.md); jevbrain WHITEPAPER N=10000 95.2% / 0.18 ms / $15.36 labeled unverified theater (55fe0abf / WHITEPAPER); jev-mail ~$0.25/20k metadata only (be3211b0); tidy 0.8 + none stay (9211939e); lkclean 70/35/60 Show (1cc18acf pin; live README still those defaults).
2	Class-table inflation (backend-agnostic ≠ multimodal omni)	PASS after lock fix. Class table only adds “placements / unofficial packaging, not new species” (references/judgment-class.md §90 bullet). No new omni/Archer/multimodal species. Rust SDK “not a new species.” jevbrain local n-gram ≠ TypeSafe Jev ≠ classifier.dev.
3	Name collisions (ego-jev authors; guard name collisions)	FAIL → fixed. methods-catalog.md Hot-click row (indexed viewport → operation+target, §65 jiangkoumo species) had ZHUBoer appended. ZHUBoer is Ego Lite observe/act + choose + __none__; the agent still plans. Now Hot-click is jiangkoumo-only with ≠ ZHUBoer (references/methods-catalog.md:125); ZHUBoer sits on Computer-use observe→score→act (references/methods-catalog.md:51). Removed ZHUBoer from SKILL Closed-vote cell (SKILL.md:133; agent still plans, not closed-vote). §9 applied-mappings namesake now says “Agent still plans (not closed-vote)” (references/applied-mappings.md:1494). Other locks hold: tpellet/hunch ≠ carldaws/hunch; jev-mail ≠ jevmail; s1_ruby ≠ hunch ≠ feelings; judgement ≠ jevql; groundedness ≠ jev-judge-bench; yuyang ≠ GodsBoy; playground ≠ HF playground.
4	Archer Hume false positives	PASS. Skip Archer / still NOT landed (research/notes.md:14656-14659; census block). None of the fifteen is treated as the 27B drop. Qwen3.8 ≠ Archer left on main.
5	Soft Noul as hard gate without envelope	PASS. Thresholds labeled sensors: jsort logits, groundedness 0.5, tidy 0.8, lkclean 70/35/60, hunch 0.5/0.65, file-search 0.8, jevbrain 0.80 (research/notes.md:14649-14653; references/formal-methods.md 1241 envelopes). Envelopes named: Ego Lite locators / __none__ / completed≠success; never-execute + exit 3; none-of-folders stay; pinned/audio/current never closed; Show anyway; ORIGIN pause-if-no-Jev + validResponse sums-to-1; crawlers verify-grounding + review queue. AUTO_ACT is not a Noul.
6	Silent fallback / soundness theater	PASS. ORIGIN: no production local-policy fallback; pause-if-no-Jev. Hard-gating AUTO_ACT / ranking-as-frequency / pasting 95.2% / 83% plumbing / 36/120 NL2Bash as class ceilings / closing pinned tabs is labeled soundness theater (research/notes.md:14649-14653). Plumbing-green called out (playground 0 promotions).
7	Duplicate sections already on main after #4/#5/#6	PASS. Fresh §90. Do not re-fold 1144 / §89. Do not re-card jiangkoumo/ego-jev / carldaws/hunch / fazlerocks/jevmail / pigeonhole. #6 is Pages landing, not a watch fold.
8	Contradictions within the PR	FAIL → fixed. Banner said “16 + product cluster” with only 15 named HIGH repos + 5-app cluster (20 sources.json adds). Now research/notes.md:14676. Also implicit_true still unsupported contradicted the pinned README/methodology (implicit_true annotations map into yes). Now implicit_true included in yes (research/notes.md:14337-14339 and skill unique fragment).

Nits (not blockers)

• SKILL trigger list still has bare undecided? abstain; Detail files use `undecided?` abstain. Pre-existing fragment-hygiene mismatch, not a new species.
• Closed-vote mapping cell still says “hot-click cousin: ego-jev” without the jiangkoumo owner (pre-existing on main). The 1241 lock is now explicit on the Hot-click row.
• stefw/lkclean pin is README SHA 1cc18acf; live default branch README is f1c740b7. Thresholds 70/35/60 and Show fail-open still match.
• yuyang README dates the 10+ call card 2026-09-20; quoted as theirs.
• Playground README scoreboard is 25 rows (7/9/8/0); VERDICT.md is 17 ideas (5 CLEARED / 8 HELD / 4 kills). Both are theirs; do not collapse them.

Parent merges. Ready for parent merge on 4ae5085.