Fold aisearchio census gaps HIGH

[24601]

User-provided 1936 HIGH fold off latest main after merged #52 (d26578f / 1843 / §129). Do not reopen or amend PR #23–#52. Do not amend #52.

Basit asked for coverage of every repo in the @aisearchio open-source Jev alternatives list. That tweet is a community census pointer. catalog ≠ endorsement. 15 GitHub links; 12 already carded or densified. This fold mints first-sightings for the three gaps.

Gaps folded (first-sighting)

1. sgoedecke/system-one (MISSING until now). Python; 20★; HEAD ebde2a2db706; README SHA d331b567e2c3; LICENSE absent. Quote theirs: batched single-token choice inference for open language models, compatible with TypeSafe. SystemOne.from_pretrained. Prefill choice_index: and constrain logits to index tokens. cache_prefix=True. TypeSafe-compatible ≠ TypeSafe replica. Namesake: sgoedecke/system-one ≠ mithalouni/system-one-open ≠ KathanModh259/system-one ≠ babybear-labs/system-one.

2. mithalouni/system-one-open (was table-only §78). Python MIT; 18★; HEAD 77f1f7cccf8a. Quote theirs: open replica of TypeSafe's Jev on Gemma 4 E2B / Gemma 3 270M (Modal). TypeSafe public eval, strict common subset (343 pairs): Jev 86.9% vs this replica 76.7% theirs. 97 ms H100 theirs. Held-out task types 74.8% theirs. HF upload pending. replica ≠ TypeSafe. §78 Benchmark Heaven row stays that board's theirs.

3. kotoba-lang/typed-decisions (was a one-link under the Hub encoder §33). Python Apache-2.0; 1★; HEAD 10d7834d3b99. Quote theirs: Jev-shaped typed-decision model on ModernBERT / DeBERTa / LLaDA-MoE with measured latency, accuracy, calibration, and training cost. DeBERTa-v3-large 0.855 / 42 ms theirs. ModernBERT-base 0.717 / 68 ms theirs. LLaDA-MoE 0.835 / 676 ms theirs. Hub com-kotobalabs/open-jev-deberta-v3-large stays §33. kotoba-lang/typed-decisions ≠ convaiinnovations/laya-typed-decisions. encoder class member not Jev replica.

Census note

The aisearchio list is an index, not a grant and not a bake-off. The 12 already-carded links stay on their original sections (laya, open-alternative-jev, openjev, jeff, kev, openjev-sglang, verdict-2.0, nimble, classifier-dev, SemIf, fast-jev-compaction, jevlike). Cousin of §77 @airesearch12 Benchmark Heaven tweet (different author, different artifact).

IDs

• notes.md §130
• composition items 497–504
• findings batch #112
• uniqueness_gate 0843+0915+jcr+0922+0940+0947+1049+1143+1248+1340+1441+1542+1643+1746+1843+1936

Open #53 (Open-Jev densify) keeps original §125 / items 425–426 / no §130 / no 497+ / no batch #112. No ID overlap.

Class recipes (not Jev-only)

Without Augustus: treat TypeSafe-compatible as a replica, 76.7% as Harbor, 0.855 as a hard gate, kotoba as Laya HF, or a 15-link list as an endorsement.

With Augustus: TypeSafe-compatible ≠ TypeSafe replica; replica ≠ TypeSafe; catalog ≠ endorsement; theirs not Harbor. Same split for any Choice/Score/Noul-style head.

Does not bump 0.5.0. Skip Archer. invented_signal: false. Soft scores ≠ hard gates. SHA move is not a replica.

Checks (this revision)

• uniqueness_gate.py ok (1936 lock 1309 chars, one substring in 21 overlays; 1843 lock retained; CHANGELOG has no dump wall)
• occupancy loop now includes range(497, 505) so deleting item 497 fails the gate (was a review FAIL: docstring claimed 497–504, loop stopped at 496)
• SKILL.md metadata.version must stay 0.5.0
• evaluate_decisions.py --self-test ok (TypeSafe-compatible ≠ TypeSafe replica / replica ≠ TypeSafe / kotoba ≠ laya / census ≠ endorsement)
• revisit_fingerprints.py --self-test ok
• SKILL.md yaml.safe_load ok

  

[24601]

Testing gate: PASS

Independent adversarial review + tests on head 61f6bcf vs origin/main d26578f (merged #52). GitHub: MERGEABLE / CLEAN. Pages build SUCCESS on pre-fix head 7e35018. This commit is uniqueness_gate occupancy + SKILL version pin only (no Pages-facing files). Do not merge from this agent.

Pushable FAIL fixed on branch (61f6bcf): uniqueness_gate.py docstring claimed composition items 497–504 exist, but the occupancy loop stopped at range(481, 497) (481–496). Deleting item 497 would have passed. Loop is now range(497, 505). SKILL.md metadata.version must stay 0.5.0. Prior fifteen 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 #52 (§129 / #111); open #53 does not steal §130 / 497+ / #112	PASS	origin/main occupies §114–§129 / composition through 496 / findings through #111. No §130 / 497–504 / #112 on main. This fold takes the next free IDs (research/notes.md ## 130; composition items 497–504; findings ## Batch #112). Unused 317–321 still absent. Only new ## heading in notes is §130. Open #53 (db616a8) densifies original §125; no §130, no 497. **, no Batch #112.
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). uniqueness_gate now refuses any other pin. CHANGELOG [0.5.0] intact; Unreleased is a short 1936 bullet + class recipe, not a bump. No uniqueness lock: dump in CHANGELOG.
3	Three first-sightings honest: sgoedecke missing→card; mithalouni promote from name-drop; kotoba ≠ laya-typed-decisions	PASS	sgoedecke/system-one absent as a GitHub card on main (tweet list had "system-one (Sean Goedecke)" only). Live REST: id 1374382889, 20★, HEAD ebde2a2db706, README git-blob d331b567e2c3, LICENSE 404, desc hash 88c6ac90a44e. README theirs: SystemOne.from_pretrained; prefill choice_index:; cache_prefix=True. TypeSafe-compatible ≠ TypeSafe replica. mithalouni/system-one-open was table-only §78 (system-one-open / 68.7 row); live 18★, HEAD 77f1f7cccf8a, README 535f33028a68, LICENSE MIT 2f6f2cf1064e, desc hash 8755f1023081. README table theirs: 343-pair 86.9% vs 76.7%; 97 ms H100; 74.8% held-out. replica ≠ TypeSafe. kotoba-lang/typed-decisions was one-link under Hub encoder §33; live 1★, HEAD 10d7834d3b99, README 4d6bbf4c4e44, Apache LICENSE 513bb5e3cb4c, desc hash 8588e6be9350. README theirs: DeBERTa-v3-large 0.855 / 42 ms; ModernBERT-base 0.717 / 68 ms; LLaDA-MoE 0.835 / 676 ms. Hub encoder card stays §33. kotoba-lang/typed-decisions ≠ convaiinnovations/laya-typed-decisions. Namesakes live: KathanModh259/system-one and babybear-labs/system-one are distinct repos.
4	uniqueness_gate + evaluate_decisions + revisit_fingerprints self-tests	PASS	All three commands exit 0 (output below). uniqueness_gate: 21 overlays; prior fifteen locks + REVISIT_LOCK byte-identical vs main; UNIQ_1936 new (1309 chars). Occupancy loop range(497, 505) (497–504 inclusive). CHANGELOG has no uniqueness lock: dump.
5	theirs on census cards; catalog ≠ endorsement; soft scores ≠ hard gates; replica ≠ TypeSafe	PASS	Evaluator helpers: typesafe_compatible_is_not_replica, replica_is_not_typesafe, kotoba_is_not_laya, census_is_not_endorsement, soft_scores_are_not_hard_gates; theirs_bench_is_not_harbor for 343-pair and 0.855/42ms. Store notes_section 130 for all three first-sightings (not a kev-style densify off §33/§78). SHA move is not a replica. Fold skips Archer rewrite.
6	Human-facing prose without em-dash walls; invented_signal false	PASS	New CHANGELOG Unreleased: 0 em dashes (en-dashes only on ID ranges / PR spans). Ecosystem / notes list items use the existing Name — gloss bullet pattern, not a dump wall. No invented_signal: true anywhere. refresh-log / notes / findings quote invented_signal: false.

IDs. Claimed notes.md §130 / composition 497–504 / findings batch #112. Verified free on d26578f. §129 stays the #52 1843 fold. Headings and items present on this branch; gate refuses stolen 317–321 and requires 497–504. Open #53 does not claim these IDs.

Adversarial (PASS after occupancy loop)

• uniqueness_gate composition loop range(497, 505) (497–504 inclusive) matches claimed IDs. Deleting item 497 now fails (proved locally).
• Densify vs new: sgoedecke missing → first-sighting. mithalouni table name-drop §78 → first-sighting card; table row stays the board. kotoba Hub encoder stays §33; GitHub training record is the new card. Pointers on §33/§78 are densify-not-sibling, not a rewrite of those PRs.
• Namesake locks live: sgoedecke ≠ mithalouni ≠ KathanModh259 ≠ babybear-labs; kotoba-lang/typed-decisions ≠ convaiinnovations/laya-typed-decisions ≠ mys/laya-typed-decisions-GGUF.
• Do not reopen Fold hourly 0042 HIGH: jev-readout, jevify, laya-jev-lab #23–Fold hourly 1843 HIGH #52. No wrappers / keys / npm / pip / uv / docker / Modal deploy recipes in the fold prose (anti-copy only).
• Densify Open-Jev: latency, providers, TREC, HF packs #53 Open-Jev densify: no §130 / no 497+ / no batch #112. Not a FAIL.

Notes, not FAILs. Star counts match live REST this pass (20 / 18 / 1). mithalouni GitHub license API is NOASSERTION; LICENSE file is MIT theirs (SHA matches). kotoba GitHub license API is also NOASSERTION; LICENSE file is Apache-2.0 theirs (SHA matches). Pages rebuild for 61f6bcf 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 1936 chars=1309 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.