Skip to content

Fold hourly 1843 HIGH - #52

Merged
24601 merged 2 commits into
mainfrom
cursor/fold-hourly-1843-high-737b
Sep 21, 2026
Merged

24601 merged 2 commits into
mainfrom
cursor/fold-hourly-1843-high-737b

Conversation

@24601

@24601 24601 commented Sep 21, 2026

Copy link
Copy Markdown
Owner

Hourly 1843 HIGH fold off latest main after merged #51 (1746 / §128). Do not reopen or amend PR #23#51. Do not amend #51.

What this hour is

Primary is a REVISIT densify of jaredpalmer/kev on the original §45 card: own-data JSONL fine-tune and --init_from warm-start LoRA/head (PR #9). Jev is the exemplar of a backend-agnostic class of categorization/scoring models, not the whole mandate. SHA move is not a replica. dabit3/jev-experiments and simota/tenbin stay densify-not-sibling.

kev own-data JSONL (densify §45)

HEAD bd058057ad0a (was 8465c4c4c294 / 1746). README SHA 84b872488915. Quote theirs: Fine-tuning on your own data. --data JSONL. --init_from warm-start LoRA/head PR #9. Kev-0.8B / 4B / 9B Qwen3.5 family. 4B new-source 0.794/0.832 theirs. 9B new-source 0.812/0.837 theirs. From-scratch on 836 support-tool decisions scored 0.33 against 0.84 for the released model; --init_from kept 0.83 there and reached 0.88 on the new domain theirs. Kev-0.5B card now points at the Qwen3.5 family. No Jev outputs were used for training. option order can change an answer. 8.2% ≥0.9 on wrong theirs (Kev-9B 7.5%).

from-scratch ≠ warm-start. JSONL labels ≠ Harbor. Qwen3.5 ≠ Archer. wire-compat ≠ logit-equiv.

Novel HIGH (first-sighting)

  • kyegomez/open-jev: unofficial research implementation with random weights. reconstruction ≠ replica. ≠ razorback16/openjev ≠ TheoLeeCJ/openjev ≠ Zefan-Cai/Open-Jev ≠ Shalimov04/open-jev.
  • jourdanlabs/assay-001: split verdict. CLINC150 ECE 0.0204 theirs. Banking77 ECE 0.0936 theirs. 8,576 responses zero type errors theirs. Not a global certificate.
  • brnyxx/jev-ra: 3-5x / ~300 ms theirs. 8.50× Wikipedia theirs. systems comparison ≠ semantic equivalence.
  • Promethe-us/awesome-jev: catalog ≠ endorsement. ≠ MrJev/awesome-jev ≠ yibie/awesome-jev.
  • ThePFMind/jev-mcp: ≠ jkudish/jev-mcp ≠ burnigtm/jev-mcp. serving substrate ≠ calibrated replica.

Remainder: minesweeper cousins (game success ≠ calibrated Noul), namenu/pi-jev-effortTheoOliveira/pi-jev, samatv256/mini-Jevr-ms/mini-jev, HF allmix vs handoff-demo, Nutlope/jev-fraud, jeffloo886/jev-notion. Skip Archer (promised_not_landed). Hub archerhume/4rcherhume HTTP 401.

IDs

  • notes.md §129
  • composition items 481–496
  • findings batch #111
  • uniqueness_gate 0843+0915+jcr+0922+0940+0947+1049+1143+1248+1340+1441+1542+1643+1746+1843

Class recipes (not Jev-only)

Without Augustus: treat a from-scratch adapter as equal to a warm start, a JSONL file as Harbor, a random-weight reconstruction as a replica, or a catalog as a grant.

With Augustus: from-scratch ≠ warm-start; JSONL labels ≠ Harbor; reconstruction ≠ replica; catalog ≠ endorsement. Same split for any Choice/Score/Noul-style head.

Does not bump 0.5.0. Skip Archer. invented_signal: false.

Checks (this revision)

Head 60b6746 remaps the 1843 archive run_digest.json off #51's IDs (§128 / 465–480 / #110) onto the free IDs (§129 / 481–496 / #111). uniqueness_gate now refuses a stale digest.

  • uniqueness_gate.py ok (1843 lock 1815 chars, one substring in 21 overlays; 1746 lock retained; CHANGELOG has no dump wall)
  • evaluate_decisions.py --self-test ok (from-scratch ≠ warm-start / JSONL labels ≠ Harbor / reconstruction ≠ replica, plus 1746 helpers)
  • revisit_fingerprints.py --self-test ok
  • SKILL.md yaml.safe_load ok
Open in Web Open in Cursor 

cursoragent and others added 2 commits September 21, 2026 01:19
kev own-data JSONL densify, from-scratch vs warm-start class claim,
assay-001 split verdict, and kyegomez reconstruction cards. Snapshot
off #50 before rebasing onto merged #51.

Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>
Archive digest still claimed §128 / 465-480 / #110 after the rebase
onto merged 1746. Point it at the free IDs §129 / 481-496 / #111 and
gate those fields in uniqueness_gate so a stale digest fails the lock.

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

24601 commented Sep 21, 2026

Copy link
Copy Markdown
Owner Author

Testing gate: PASS

Independent adversarial review + tests on head 60b6746 vs origin/main 26915ca (merged #51). GitHub: MERGEABLE / CLEAN. Pages build SUCCESS on pre-fix head fea5c24. This commit is archive digest + uniqueness_gate only (no Pages-facing files). Do not merge from this agent.

Pushable FAIL fixed on branch (60b6746): research/archive/hourly/2026-09-21T00/run_digest.json still claimed #51 IDs (notes_section 128 / composition 465-480 / findings_batch 110) after the rebase remap. Now §129 / 481-496 / #111. uniqueness_gate now refuses a stale 1843 digest. Prior fourteen uniqueness locks + REVISIT_LOCK are byte-identical vs origin/main. CHANGELOG [0.5.0] identical vs main.

# Check Result Evidence
1 IDs free vs main after #51 (§128 / #110) PASS origin/main occupies §114–§128 / composition through 480 / findings through #110. No §129 / 481–496 / #111 on main. This fold takes the next free IDs (research/notes.md ## 129; composition items 481–496; findings ## Batch #111). Unused 317–321 still absent. Only new ## heading in notes is §129. Digest remapped to match.
2 SKILL metadata stays 0.5.0; no release bump PASS YAML metadata.version: 0.5.0; typesafe_skill: v0.5.7 @ 65a39f3 (identical vs main). CHANGELOG [0.5.0] intact; Unreleased is a short 1843 bullet + class recipe, not a bump. No uniqueness lock: dump in CHANGELOG.
3 Primary REVISIT densify kev on original §45; SHA move ≠ replica; --data JSONL / --init_from theirs; no sibling first-sighting PASS Densify card appended on §45 (### Since last look (2026-09-21T00 hourly 1843)). Store notes_section stays 45. Helper densify_original_ids refuses moving kev off 45. Live HEAD still bd058057ad0a / README git-blob 84b872488915 / desc hash 47fc60928c90. Live README has Fine-tuning on your own data; --data JSONL; --init_from; 0.33 / 0.84 / 0.83 / 0.88; 0.794/0.832; 0.812/0.837; No Jev outputs were used for training theirs. dabit3 / tenbin densify under §129 with no dedicated ## headings. No sibling first-sighting for kev.
4 uniqueness_gate + evaluate_decisions + revisit_fingerprints self-tests PASS All three commands exit 0 (output below). uniqueness_gate: 21 overlays; prior fourteen locks + REVISIT_LOCK byte-identical vs main; UNIQ_1843 new (1815 chars). CHANGELOG has no uniqueness lock: dump.
5 theirs on novel HIGH; Archer still promised_not_landed PASS Live kyegomez README: from-first-principles / unofficial / random weights. assay-001: CLINC150 0.0204 / Banking77 0.0936 / 8,576 theirs. jev-ra GitHub description 3-5x / ~300 ms; README ~300 ms and 8.50. Promethe-us curated list. ThePFMind jev_evaluate / jev_route. Description sha256[:12] match kev 47fc60928c90, kyegomez 6913694477a7, assay d602297bea69, jev-ra 0243f197153c, ThePFMind b9bc5fd053eb. Remainder 34/34 cards theirs. Fold skips Archer rewrite. Hub archerhume/4rcherhume live HTTP 401 this pass. Item 496 / notes skip: promised_not_landed held.
6 Human-facing prose without em-dash walls; invented_signal false PASS New CHANGELOG / recipes.html added lines: 0 em dashes (en-dashes only on ID ranges / PR spans). Densify headings use the existing Since last look — repo pattern. No invented_signal: true anywhere. refresh-log / notes / findings / digest quote invented_signal: false.

IDs. Claimed notes.md §129 / composition 481–496 / findings batch #111. Verified free on 26915ca. §128 stays the #51 1746 fold. Headings and items present on this branch; gate refuses stolen 317–321 and requires 481–496.

Adversarial (PASS after digest remap)

  • uniqueness_gate composition loop range(481, 497) (481–496 inclusive) matches claimed IDs. Deleting item 481 would fail.
  • Densify vs new: kev densifies §45. dabit3 / tenbin densify existing mentions (store notes_section 129; no sibling ## headings). kyegomez / assay-001 / jev-ra / Promethe-us / ThePFMind are first sightings (notes_section 129).
  • Namesake locks live: Promethe-us/awesome-jev ≠ MrJev ≠ yibie; kyegomez/open-jev ≠ razorback16 ≠ TheoLeeCJ ≠ Zefan-Cai ≠ Shalimov04; ThePFMind/jev-mcp ≠ jkudish ≠ burnigtm; namenu/pi-jev-effort ≠ TheoOliveira/pi-jev; samatv256/mini-Jev ≠ r-ms/mini-jev; comoc/jev-minesweeper ≠ EnesYilmazcode/JevMinesweeper.
  • Do not reopen Fold hourly 0042 HIGH: jev-readout, jevify, laya-jev-lab #23Fold hourly 1746 HIGH #51. No wrappers / keys / npm / pip / uv / docker install recipes in the fold prose (anti-copy only).

Notes, not FAILs. Star-noise after snapshot: kev 1051★ vs fold 1033 (SHA still bd058057ad0a); kyegomez 2★ vs 1. jev-ra README uses ~300 ms / 8.50; the 3-5x string is on the GitHub description, not a README substring this pass. Promethe-us README is a curated list; not affiliated is fold gloss, not a README substring. dabit3 / tenbin have no original ## N so store notes_section 129 is densify-under-this-hour, not a stolen kev-style original id. Pages rebuild for 60b6746 was still pending when this comment posted.

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 1643 chars=2193 1746 chars=1902 1843 chars=1815 revisit chars=447 overlays=21 revisit_overlays=5
self-test ok
revisit-fingerprints self-test ok

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

@24601
24601 merged commit d26578f into main Sep 21, 2026
1 check passed
@24601
24601 deleted the cursor/fold-hourly-1843-high-737b branch September 21, 2026 01:34
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