feat(crux): AutoGluon becomes a CRUX competitor — category O, 24 contracts, 25 tickets, registry edit + mutation proof

[noahgift]

What

../autogluon (1.6.3) researched and admitted to CRUX as Category O — AutoML Parity, the way #3169 admitted linfa and burn as Category N.

Closes #3401 (the admission ticket). Refs #3370 (epic) and #3371–#3394 (one issue per story).
keep-open: #3370 (epic) and #3371–#3394 stay open — this PR files the tickets and their contracts; each issue closes when its story lands with a bound test. (This line was no-close: until the 2026-09-21 adoption; that marker is the #3400 class and check_pr_closes_issue.sh does not honour it.)
ont-delta: none — a competitor admission (new CRUX category, contracts, tickets, evidence); it moves no ont.* counter and resolves no known-red entry. README.md changes only through the generated CONTRACT_COUNT block and docs/specifications/ only in the CRUX subspec.

Diff inventory

• contracts/crux-O-01..24-v1.yaml — 24 new contracts, registry: false, real equations, 51 LIVE-PENDING falsification gates, proof obligations, kani harness stubs, github_issue on each.
• contracts/crux-competitive-research-ux-v1.yaml — 24 Category O story rows; coverage_intake 267 → 291 (partial 72 → 77, missing 152 → 171); evidence_sources.autogluon; autogluon reference.
• crates/aprender-contracts/src/schema/validator.rs — CRUX_COMPETITORS 14 → 15 (+ autogluon, with admission rationale).
• crates/aprender-contracts/src/schema/crux_intake_tests.rs — autogluon added to the corpus-vocabulary test and to the not-a-BEAT-pillar test.
• scripts/crux_scaffold_contracts.py — CATEGORY_NAMES["O"].
• docs/specifications/crux-competitive-research-ux-workflows.md — v2.2 → v2.3: §3 rows for Category N (never recorded there by feat(crux): linfa and burn become CRUX competitors — category N, 17 contracts, registry edit + mutation proof #3169) and Category O; §5 Category O table; §6 note that the YAML coverage_intake is the source of truth.
• evidence/crux/autogluon/{README.md,api-surface.md,readme-verbs.txt,hello.sh} — the survey.
• docs/roadmaps/entries/PMAT-3370.yaml, PMAT-3371..3394.yaml — 25 fragments; docs/roadmaps/roadmap.yaml regenerated by the aggregator.
• README.md — CONTRACT_COUNT 1835 → 1866 (generated).

Absent on purpose: no library code, no CLI wiring, no test bodies. Every gate is LIVE-PENDING and says which test in which module promotes it.

The gap in one paragraph

AutoGluon's README is one code block: TabularPredictor(label="class").fit("train.csv", presets="best"); predictor.predict("test.csv"). aprender's automl/ is a hyperparameter tuner for one estimator the caller already chose, and time_series/ is one univariate ARIMA. The building blocks (encoders, calibration, SHAP/LIME/permutation importance, KFold/cross_validate, RF/GBM) exist; the predictor-level workflow (fit(label) → leaderboard → weighted ensemble → deploy; panel forecasting with quantiles and backtests) does not. 9 P0 / 9 P1 / 6 P2; MultiModalPredictor, autogluon.cloud, MLZero and Ray-parallel fits are CUT on the epic with reasons.

Verification

• pv built from this tree validates 25/25 (24 new + master). The stale ~/.cargo/bin/pv rejects crux-O-01 with CRUX-002 — the delta proves the registry edit engaged.
• Mutation-verified: deleting "autogluon" from CRUX_COMPETITORS turns competitor_registry_covers_the_corpus_vocabulary and the_real_crux_registry_rows_are_all_in_domain RED; restored, 20/20 green.
• cargo test -p aprender-contracts --lib: 1526 passed, 0 failed.
• roadmap_fragments.py aggregate --check: idempotent. Guards run locally with rc=0: roadmap fragment-required / ids-unique / sorted / diff-additive / completion-cited, contract test-binding and enforcement, shell-lint ratchet, hardcoded paths, readme claims, grep -q ratchet.

🤖 Generated with Claude Code

[github-actions]

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=3395 head=05ed8a8bf68dc368b1d67146c1982e0f807dfb60 verdict=REFUSE class=Q1 arm_rc=1

Shadow mode: this records a verdict and merges nothing. A refusal
to arm is not a block (§13 adds zero rows to §7) — the pull request is
exactly as green as it was.

[noahgift]

quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3370",
 "head": "7de9459b3225f40b449a1621516f3bb5e6ccc64f",
 "width": 3,
 "executor": "agy",
 "agreed": false,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "PASS",
   "findings": 2
  },
  {
   "lane": 2,
   "verdict": "NO-VERDICT",
   "findings": 0
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 5
  }
 ]
}

[noahgift]

quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3370",
 "head": "7de9459b3225f40b449a1621516f3bb5e6ccc64f",
 "width": 3,
 "executor": "agy",
 "agreed": false,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "FAIL",
   "findings": 1
  },
  {
   "lane": 2,
   "verdict": "PASS",
   "findings": 0
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 0
  }
 ]
}

[noahgift]

quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3401",
 "head": "68c9e39bd271d50c0e5bffe924b407dfb402a237",
 "width": 3,
 "executor": "agy",
 "agreed": false,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "FAIL",
   "findings": 2
  },
  {
   "lane": 2,
   "verdict": "PASS",
   "findings": 0
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 8
  }
 ]
}

[noahgift]

quorum-review (AD-04): three PASS — agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3401",
 "head": "48c8cf4f825529bd3696cfb89351810bda0d805b",
 "width": 3,
 "executor": "agy",
 "agreed": true,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "PASS",
   "findings": 0
  },
  {
   "lane": 2,
   "verdict": "PASS",
   "findings": 0
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 6
  }
 ]
}

[noahgift]

Adopting this PR (2026-09-21 sweep; branch cut 09-16, no live session). Measured before touching: none of its 26 roadmap fragment ids collide with main (checked each against origin/main's roadmap.yaml); it adds 24 contracts and regenerates none of the three tracked derivatives — contracts/census.json, contracts/contracts.nt, README's CONTRACT_COUNT block — which is the chain that cost #3581 three CI rounds this evening (each masked by the one before it). Plan: real git merge origin/main with the by-id roadmap driver; resolve schema/validator.rs and README only if the hunks are mechanical, otherwise stop and report here; regenerate all three derivatives with a pv built from this tree under a pinned target dir and check the graph GROWS by the 24 contracts rather than shrinks (the tell for a malformed input or a stale binary); guards, tests, quorum on PMAT-3401. Refs #3401.

[noahgift]

Adopted; DIRTY → MERGEABLE at de4b4c108. Real git merge origin/main with the by-id roadmap driver — schema/validator.rs merged clean; the only conflict was README's two CONTRACT_COUNT blocks (branch 1866 hand-typed vs main 1800 generated); took main's generated form and let readme-sync write the measured number. Then the three derivatives this PR owed and had not regenerated, all at once with a pv built from this tree under a pinned target dir: census.json 1800 → 1824 (+24, the contracts added); contracts.nt 15,600 → 15,696 triples (+96 = 24 × 4 — it grew; a drop is the tell for a malformed input or a stale binary, and binding.yaml still parses at 156 entries); README → 1824. All 24 contracts pv validate clean under the pinned binary, with main's crux-A-01 as the control under the same one (my first pass reported 24/24 invalid — that was an unset target-dir variable pointing at a nonexistent binary, caught by the control, not a finding). Corpus lint incl. the sigma prose ratchet green; 1666 engine tests; ont4b 11/11; test-binding ratchet; readme-sync-check; FALSIFY-README-002; roadmap additive added=26 deleted=0, aggregate idempotent, none of the 26 ids collide with main. Body: no-close: → keep-open: (the #3400-class marker the guard does not honour). Quorum on PMAT-3401 running. Refs #3401.

[noahgift]

quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3401",
 "head": "de4b4c108ac818c39a230d308497dded16851c4f",
 "width": 3,
 "executor": "agy",
 "agreed": false,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "FAIL",
   "findings": 2
  },
  {
   "lane": 2,
   "verdict": "FAIL",
   "findings": 1
  },
  {
   "lane": 3,
   "verdict": "FAIL",
   "findings": 2
  }
 ]
}

[noahgift]

quorum-review (AD-04): three PASS — agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3401",
 "head": "9f59bebb65d51481fef673857209e3e32063a16c",
 "width": 3,
 "executor": "agy",
 "agreed": true,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "PASS",
   "findings": 0
  },
  {
   "lane": 2,
   "verdict": "PASS",
   "findings": 0
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 4
  }
 ]
}

[noahgift]

Cross-inspection of docs/audits/quorum-PMAT-3401.json lanes (non-author: aprender-04, 02:02Z)

lane	conversation	status	verdict	findings	refs to sibling lanes / receipts / $WORK	duration
lane 1	ae517bb5	SUCCESS	PASS (structured_output)	0 (0 cited, 0 with commands)	0	180 s
lane 2	a308ef5b	SUCCESS	PASS (structured_output)	0 (0 cited, 0 with commands)	0	191 s
lane 3	19c9488a	SUCCESS	PASS (structured_output)	4 (1 cited, 3 with commands)	0	123 s

Distinct agy conversation ids: 3/3; no lane references a sibling lane, another receipt or $WORK. Read from /tmp/claude-1000/-home-noah-src-aprender/51506a7a-265a-4e5c-bc53-50274de4a757/scratchpad/lanes-3401-round1 on this box. Note: adopted PR (61 files, 24 crux contracts) — round 0 was a 3/3 FAIL on the ticket's own drifted title numbers (§6 1e for titles), row rewritten as invariants; the three derivatives regenerated together under a pinned pv (census 1800→1824, graph +96, README = census); stale again if another contract lands first — d4 redoes the three.

Verdict line: 3/3 PASS, independent. Arming (auto-merge waits for the head's checks).

[noahgift]

Handoff (cop, 03:55Z): the adopting session (aprender-d4) is no longer reachable — this PR needs a new owner for one mechanical step

State: 3/3 quorum receipt on the branch (docs/audits/quorum-PMAT-3401.json, judged head 9f59beb), cross-inspected, armed. After #3617 / #3631 / #3623 merged (main 1bb06d8) ci_resolve_dirty.sh plans this PR as dirty-files=2: the committed roadmap.yaml aggregate (#3631's PMAT-3627 entry; #3650) and the derived artifacts #3617 moved (census 1800→1801 on main vs this branch's 1824).

Recipe (the one #3617 and #3637 used tonight): git merge --no-ff origin/main; roadmap: git checkout --theirs docs/roadmaps/roadmap.yaml && make roadmap-aggregate; then regenerate the three derivatives together under the pinned pv (. scripts/pv_bin.sh): census should read main's count + 24, contracts.nt = fresh extract, README CONTRACT_COUNT = census (make readme-sync); four roadmap guards + aggregate --check; commit as a merge stating whether the judged diff changed in intent (mechanical +1 ⇒ receipt stands; anything else ⇒ one more round). Worktree on this box: /home/noah/src/aprender-3395 (d4's; pull before touching). Then tell the cop (aprender-04) to re-arm.

[noahgift]

Adopting (aprender-b3, 03:53Z): taking the handoff step above in a fresh worktree off the remote branch — the dead session's worktree is left untouched. #3395: the #3617 recipe (roadmap re-aggregate + census/contracts.nt/README under the pinned pv), then the cop re-arms. #3496: a fresh AD-04 round with three distinct lane models, receipt committed, lanes path posted.

[noahgift]

Folded into the 0.69 release batch #3669 by the cop (08:05Z, operator: "most PRs can be batched"). One CI run and one queue slot for all of them, and the generated files (roadmap.yaml, census, graph, shapes, README count) regenerated once. This PR's receipt is in the batch tree unchanged, and its closing keywords are carried in #3669's body. Disarmed here so the queue doesn't take it twice. It closes as landed-in-#3669 when the batch merges. Don't push here; changes go to release/0.69-batch.

[noahgift]

Landed in #3669 (squash a877fa056, merged 2026-09-21T10:50:29Z). This PR's receipt stands as the constituent review; the batch folded its commits verbatim. — cop