Fold hourly 1542 HIGH

[24601]

Hourly 1542 HIGH fold into Augustus on a fresh branch off latest main. Does not reopen or amend PR #23–#48. Does not bump 0.5.0. Does not reply into finished 1441 agents or merged #48.

What this fold adds

research/notes.md §126, composition items 433–448, findings batch #108. uniqueness_gate now checks the 1542 lock as one consecutive substring across the 21 overlays, on top of 0843 + 0915 + jcr + 0922 + 0940 + 0947 + 1049 + 1143 + 1248 + 1340 + 1441.

Primary lens (REVISIT densify): TypeLLM/TypeLLM README 3k→12k B (thinking budget, batch 5.8x theirs, Constrained AR ≠ calibrated Noul) and jaredpalmer/kev family README (new-source theirs, 8.2% ≥0.9 on wrong theirs, option order can change an answer).

Revisit densify (keep original section ids):

• TypeLLM/TypeLLM HEAD 6a48f9f1e623 (was c4b03ba9e792). README SHA dbdc1f193537. Quote theirs: thinking=True/False per-field budget. type safety does not guarantee factual accuracy. Sequential 9.35 s vs batch 1.61 s at K=16 Boolean fields, 5.8x theirs. Constrained AR ≠ calibrated Noul. Qwen/Qwen3.8-27B ≠ Archer. Densify §113, not a sibling first sighting.
• jaredpalmer/kev HEAD b339f446a0ef (was 75cc15ddb8e2). README SHA 86b0a19909f3. Live 970★ vs watch 964 is star-noise. Quote theirs: Questions share the input text but cannot read each other. No Jev outputs were used for training. 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. 8.2% ≥0.9 on wrong theirs. option order can change an answer. Qwen3 ≠ Archer. Densify §45 / §98.
• TheoOliveira/pi-jev HEAD 549c2bfa2492 densify §42 (watch listed novel HIGH, already catalogued). Quote theirs: Fails closed safely: unjudged tools do not blindly activate. JEV_THRESHOLD 0.65 still soft. routing ≠ permission.

Novel HIGH (selected):

• harshwasan/jev-sentinel 8★: fail closed never auto-allows. harshwasan/jev-sentinel ≠ leepokai/jev-guard. rh-guard owns primary gates.
• 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. does not execute. 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.
• hf:rAVEUK/open-jev-deberta-v3-large encoder class member not Jev replica. hf:p-yan/laya-quanto and hf:Gtrkrsk/laya serving substrate ≠ calibrated replica. p-yan/laya-q8 and q4 Hub HTTP 401.

Remainder playgrounds and name-match demos stay short cards. Do not mint sibling first sightings for catalogued repos.

Class table rows (TypeLLM / kev family) and Pages recipe cards densified: Constrained AR vs Noul, kev isolation, fail polarity, classifier ≠ authorizer. Recipes show class benefits, not Jev-only.

Honesty

Third-party benches stay theirs. No equivalence guarantees. SHA move is not a replica. catalog ≠ endorsement. Skip Archer. Hub archerhume/4rcherhume HTTP 401. Archer still promised_not_landed. invented_signal: false.

Tests

python3 .agents/skills/augustus/scripts/uniqueness_gate.py
python3 .agents/skills/augustus/scripts/evaluate_decisions.py --self-test
python3 research/revisit_fingerprints.py --self-test

All three pass on 65faa3f. Fingerprint store now keeps full Hub SHAs for rAVEUK / p-yan/laya-quanto / Gtrkrsk/laya (lock still quotes the 12-char prefixes). Evaluator additions: Constrained AR ≠ Noul; fail-closed never auto-allows; fail-open uncertainty means RUN; routing threshold still soft (0.65 / 0.90 / 0.7); classifier ≠ authorizer; estimates not Harbor.

  

[24601]

Testing gate: PASS

Independent adversarial review + tests on head 88d9872 vs origin/main f40139c (merged #48). GitHub: MERGEABLE / CLEAN. Pages build SUCCESS (run 35541532646). Do not merge from this agent.

Pushable FAILs fixed on this head: (1) revisit_fingerprints.py densify helper now locks github:TheoOliveira/pi-jev on notes_section 42 (moving it to 126 fails --self-test); (2) esinocchi description is GitHub-null, store now null not empty-string sha256 e3b0c44298fc; (3) CONTRIBUTING uniqueness_gate line lists 1542.

#	Check	Result	Evidence
1	IDs free vs main after #48 (§125 / #107)	PASS	origin/main occupies §114–§125 / composition through 432 / findings through #107. No §126 / 433–448 / #108 on main. This fold takes the next free IDs (research/notes.md ## 126; composition items 433–448; findings ## Batch #108). Unused 317–321 still absent. No remap. Only new ## heading in notes is §126. uniqueness_gate composition loop this hour is range(433, 449) (433–448 inclusive); deleting item 433 fails.
2	SKILL metadata stays 0.5.0; no release bump	PASS	YAML metadata.version: 0.5.0; typesafe_skill: v0.5.7 @ 65a39f3. README Current 0.5.0. marketplace / CITATION.cff / SECURITY 0.5.x unchanged vs main. CHANGELOG [0.5.0] intact; Unreleased is a short 1542 bullet + class recipe, not a bump.
3	REVISIT densify on original ids (TypeLLM, kev, pi-jev §42); Constrained AR ≠ calibrated Noul; classifier ≠ authorizer; theirs	PASS	Densify cards on original sections: TypeLLM §113 (### Since last look (2026-09-20T21 hourly 1542)), kev §45, pi-jev §42 item 9. Store notes_section 113 / 45 / 42. Helper refuses moving those ids. No sibling first-sighting headings. Live README at pinned SHAs: TypeLLM thinking=True/False per-field budget; type safety does not guarantee factual accuracy; Sequential 9.35 s vs batch 1.61 s, 5.8x theirs. kev: share the input text / can't read each other; No Jev outputs were used for training; 4B new-source 0.790/0.806 theirs; 8.2% ≥0.9 on wrong theirs; option order can change an answer. pi-jev: Fails closed safely … unjudged tools. Evaluator: constrained_ar_is_not_noul / classifier_is_not_authorizer / estimates_are_not_harbor.
4	uniqueness_gate includes 1542 + prior locks; evaluate_decisions + revisit_fingerprints	PASS	All three commands exit 0 (output below). uniqueness_gate: 21 overlays; prior eleven locks + REVISIT_LOCK byte-identical vs main; UNIQ_1542 new (2015 chars). CHANGELOG has no uniqueness lock: dump. Novel HIGH first sightings notes_section 126: jev-sentinel / tool-routers / leanest / jevals / MrJev / jev-firewall / jev-codex-approval / HF encoder-quanto. Live description sha256[:12] match TypeLLM 4e7869c96688, kev 8f68dcc43f12, pi-jev 2155badd27d8, sentinel 3b4ee158cdc9. Hub rAVEUK 7beac18983e6 / p-yan/laya-quanto 070a6e609763 / Gtrkrsk/laya f6029af5253c match. README git-blob SHAs match dbdc1f193537 / 86b0a19909f3 / a465d819f32b.
5	Archer still promised_not_landed; Hub 401; Qwen3.8-27B ≠ Archer	PASS	Fold skips Archer rewrite. Hub archerhume/4rcherhume live HTTP 401 this pass. p-yan/laya-q8 and q4 HTTP 401. Qwen/Qwen3.8-27B ≠ Archer. Item 448 / notes skip: promised_not_landed held.
6	Human-facing prose without em-dash walls; recipes class not Jev-only	PASS	New CHANGELOG / recipes / §126 how-to-apply uses slashes and periods. Added em-dashes: since-last-look heading house style ( — repo) plus pre-existing SKILL protocol lines that grew by lock fragments. No new em-dash dump walls. Pages cards: Constrained AR / kev isolation / fail polarity / classifier ≠ authorizer. CHANGELOG recipe: same split for any Choice/Score/Noul-style head, not only hosted Jev.

IDs. Claimed notes.md §126 / composition 433–448 / findings batch #108. Verified free on f40139c. §125 stays the #48 1441 fold. Headings and items present on this branch; gate refuses stolen 317–321 and requires 433–448.

Adversarial (PASS, no further fix)

• uniqueness_gate composition loop as above (88d9872).
• Densify vs new: TypeLLM densifies §113. kev densifies §45. pi-jev densifies §42. sentinel / routers / leanest / jevals / MrJev / firewall / jev-codex-approval / HF encoder-quanto are first sightings. No sibling first-sighting headings.
• Live README quotes match 5.8x, 0.790/0.806, 8.2%, 76/81, Fail open uncertainty means RUN, never auto-allows, classifier become the authorizer, estimates until jevals bench.
• Do not reopen Fold hourly 0042 HIGH: jev-readout, jevify, laya-jev-lab #23–Fold hourly 1441 HIGH #48. No wrappers / keys / npm / pip / docker run recipes in the fold prose (anti-copy only).

Notes, not FAILs. Watch prompt said TypeLLM README 3k→12k B; last-look SHA c4b03ba9 was already 12367 B and already contained thinking / 5.8x; this hour is banner/intro +293 B (git blob dbdc1f193537). kev live README says "can't read each other"; lock says "cannot". pi-jev quote is condensed from "if Jev is unreachable or unconfigured, tool routing does not blindly activate unjudged tools". TypeLLM live 17★ / kev 975★ vs fold 16 / 970 (star-noise). openlayer-ai/jevals HEAD has moved after this snapshot (live ed52b9c / pushed 22:08:27Z; store a38a971 / 21:55:09Z). Evaluator module docstring still ends at 1340 coverage-at-error-budget; 1441/1542 helpers are in --self-test. CONTRIBUTING "do not reopen #23–#45" sentence is older and stale, not this hour's ID claim.

uniqueness-gate ok
0843 chars=2423 0915 chars=1132 jcr chars=1499 lock0922 chars=1292 0940 chars=1350 0947 chars=2350 1049 chars=2162 1143 chars=2188 1248 chars=2226 1340 chars=1499 1441 chars=1635 1542 chars=2015 revisit chars=447 overlays=21 revisit_overlays=5
self-test ok
revisit-fingerprints self-test ok

Ready for parent review. PR stays open (draft). Do not merge from this agent.