Skip to content

Fold hourly 1441 HIGH - #48

Merged
24601 merged 1 commit into
mainfrom
cursor/fold-hourly-1441-high-aa6d
Sep 20, 2026
Merged

24601 merged 1 commit into
mainfrom
cursor/fold-hourly-1441-high-aa6d

Conversation

@24601

@24601 24601 commented Sep 20, 2026

Copy link
Copy Markdown
Owner

Hourly 1441 HIGH fold into Augustus on a fresh branch off latest main. Does not reopen or amend PR #23#47. Does not bump 0.5.0. Does not reply into finished 1340 agents or merged #47.

What this fold adds

research/notes.md §125, composition items 417–432, findings batch #107. uniqueness_gate now checks the 1441 lock as one consecutive substring across the 21 overlays, on top of 0843 + 0915 + jcr + 0922 + 0940 + 0947 + 1049 + 1143 + 1248 + 1340.

Primary lens (REVISIT densify): razorback16/openjev STE README rewrite. vLLM NVIDIA + MLX Apple Silicon. Codiv hosted free endpoint. dual /v1/systemone + /v1/chat/completions. serving-port densify.

Revisit densify (keep original section ids):

  • razorback16/openjev HEAD cddbd962c88a (was 6e91dfc031bc). Quote theirs: two backends serve the same /v1/systemone. /v1/chat/completions needs vLLM. The MLX backend answers 501. Hosted for free on Codiv. Independent; not affiliated with TypeSafe AI. dual serving is not generate. Hosted Codiv ≠ TypeSafe. Error contract is not a Noul. wire-compat ≠ logit-equiv. SHA move is not a replica.

Novel HIGH (selected):

  • hr98w/jev-visual 167★: Apple Silicon visual candidate scoring. Quote theirs: Candidate probabilities are relative to supplied options, not correctness estimates. 37.30s → 2.40s at 64 decisions theirs. Breakout 9 bricks 6 returns 2 lives theirs. hr98w/jev-visual ≠ sseanliu/Jev-Vision.
  • jkudish/jev-mcp 156★: ten MCP tools. Quote theirs: The recommendation is advisory. The server never blocks on its own. TypeSafe CLERC 5% to 18% theirs. jkudish/jev-mcp ≠ burnigtm/jev-mcp. Gate cousins stay measurement notes; rh-guard owns primary gates.
  • zhengxuyu/litjev: off-the-shelf Qwen decision layer. Quote theirs: Probabilities are not calibrated by default. Qwen/Qwen3.8-27B ≠ Archer. zhengxuyu/litjev ≠ alexwestco/llm-to-jev.
  • Zefan-Cai/Open-Jev + Hub ZefanCai/Open-Jev-{2,9}B: 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.
  • hf:AndyInQtr/laya-coreai: CoreML serving substrate ≠ calibrated replica. AndyInQtr/laya-coreai ≠ mizorewww/laya-coreml.

pngwn/open-jev, Praveenrajus/jev-bench (Hub HTTP 401), jevify-gemma4*, and bonzi* stay remainder. Do not mint sibling first sightings.

Class table rows (Judgment-model class / Open weights) and Pages recipe cards densified. 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.

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 this revision. Evaluator additions: dual serving is not generate; Hosted Codiv ≠ TypeSafe; candidate probabilities relative not correctness; MCP recommendation is advisory; LoRA ≠ RLCD replica; pass-min 0.8 still soft.

Open in Web Open in Cursor 

Densify razorback16/openjev (STE README: vLLM NVIDIA + MLX Apple Silicon,
Codiv hosted endpoint, dual /v1/systemone + /v1/chat/completions). Fold
jev-visual, jev-mcp, litjev, Open-Jev LoRA, jeq, and laya-coreai serving
honesty. uniqueness_gate 1441 / notes §125 / items 417-432 / batch #107.
Does not bump 0.5.0.

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>
@24601

24601 commented Sep 20, 2026

Copy link
Copy Markdown
Owner Author

Testing gate: PASS

Independent adversarial review + tests on head 1a1ecfa vs origin/main fea7c0b (merged #47). GitHub: MERGEABLE / CLEAN. Pages build SUCCESS (run 35537376087). Do not merge from this agent.

No pushable FAIL. uniqueness_gate composition loop this hour is range(417, 433) (417–432 inclusive) and matches the docstring. Deleting the 1441 algebra block would fail, unlike the 1248 hole on 8d755c1.

# Check Result Evidence
1 IDs free vs main after #47 (§124 / #106) PASS origin/main occupies §114–§124 / composition through 416 / findings through #106. No §125 / 417–432 / #107 on main. This fold takes the next free IDs (research/notes.md ## 125; composition items 417–432; findings ## Batch #107). Unused 317–321 still absent. No remap. Only new ## heading in notes is §125.
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 1441 bullet + class recipe, not a bump.
3 Primary REVISIT densify openjev on original section; dual serving ≠ generate; Hosted Codiv ≠ TypeSafe; SHA move ≠ replica PASS Densify card appended on §75 (### Since last look (2026-09-20T20 hourly 1441)). Store notes_section stays 75. Live README at pinned SHA a5943415cb92 / HEAD cddbd962c88a: two backends serve the same /v1/systemone; /v1/chat/completions needs vLLM; MLX answers 501; Hosted for free on Codiv; Independent, not affiliated with TypeSafe AI. Evaluator: dual_serving_is_not_generate / hosted_codiv_is_not_typesafe.
4 uniqueness_gate + evaluate_decisions + revisit_fingerprints; theirs on novel HIGH PASS All three commands exit 0 (output below). uniqueness_gate: 21 overlays; prior ten locks + REVISIT_LOCK byte-identical vs main; UNIQ_1441 new (1635 chars). CHANGELOG has no uniqueness lock: dump. Helper still refuses moving 1248/1340 densify stores off original section ids (openjev §75, kev §45, …). Novel HIGH first sightings notes_section 125: jev-visual / jev-mcp / litjev / Open-Jev / jeq / laya-coreai. Live README quotes match 37.30s → 2.40s, 9 bricks 6 returns 2 lives, CLERC 5% to 18%, 2B 94.71% / 9B 97.54% hard test, OOD 86.02% / 91.97%, 80,816 rows, pass-min 0.8.
5 Qwen3.8-27B ≠ Archer; Archer still promised_not_landed PASS Fold skips Archer rewrite. Hub archerhume/4rcherhume live HTTP 401 this pass. Qwen/Qwen3.8-27B ≠ Archer (litjev default checkpoint). Item 432 / notes skip: promised_not_landed held.
6 Human-facing prose without em-dash walls PASS New CHANGELOG / recipes / §125 how-to-apply prose uses slashes and periods. Added em-dashes: since-last-look heading house style ( — repo) plus two pre-existing SKILL table/protocol lines that grew by lock fragments. No new em-dash dump walls.

IDs. Claimed notes.md §125 / composition 417–432 / findings batch #107. Verified free on fea7c0b. §124 stays the #47 1340 fold. Headings and items present on this branch; gate refuses stolen 317–321 and requires 417–432.

Adversarial (PASS, no fix)

  • uniqueness_gate composition loop as above (1a1ecfa).
  • Densify vs new: openjev densifies §75. jev-visual densifies the existing judgment-class region-Choice card and takes §125 as the measured first sighting (§15 still only a one-liner). No sibling first-sighting headings.
  • Fingerprint description_hash values match live GitHub description sha256[:12] (openjev 3da60ee053e0, jev-visual 41bb0f55eef8, jev-mcp 09c9fcf00b33, litjev 629fde7f9b44, jeq a316b10cbb88). Open-Jev description is null as stored. Hub ZefanCai/Open-Jev-2B 0c7aa498b162 / 9B 47e966881e48 match. laya-coreai last-look 481043ac132c existed at 2026-09-20T20:00:48Z.
  • Do not reopen Fold hourly 0042 HIGH: jev-readout, jevify, laya-jev-lab #23Fold hourly 1340 HIGH #47. No wrappers / keys / npm / pip / docker run recipes in the fold prose (anti-copy only).

Notes, not FAILs. Evaluator theirs_bench_is_not_harbor is an honesty lock (always theirs). jeq census 3★; live REST this pass 2★ (ephemeral). hf:AndyInQtr/laya-coreai HEAD has moved after this snapshot (live d15918623256 / lastModified 20:57:03Z); store last-look is 481043ac132c / 20:00:48Z. Evaluator module docstring still ends at 1340 coverage-at-error-budget; 1441 helpers are in --self-test. CONTRIBUTING uniqueness_gate line now lists 1340+1441; the "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 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.

@24601
24601 marked this pull request as ready for review September 20, 2026 21:08
@24601
24601 merged commit f40139c into main Sep 20, 2026
1 check passed
@24601
24601 deleted the cursor/fold-hourly-1441-high-aa6d branch September 20, 2026 21:08
@24601 24601 mentioned this pull request Sep 20, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants