spec(port): port specs/port/trios/crates/trios-igla-race/src/lessons.t27 (Closes #9413)

[dmitrii-f-t27]

Summary

Port gHashTag/trios:crates/trios-igla-race/src/lessons.rs (at 8b229e9489ee, 260+ lines) as specs/port/trios/crates/trios-igla-race/src/lessons.t27 — Roadmap Stage 6 (goal #4548), target t27c gen-rust. Closes #9413.

The original is the Failure Memory: automatic lesson generation from pruned trials.

Per the issue's rule ("Port the decision, not the plumbing"), all five listed functions are ported under their names as pure functions (no transports — the database round-trips are the callers' plumbing) with 8 tests (48 runtime asserts, 0 vacuous, 100% function coverage), including the original's own high-lr and small-model tests:

• LessonType_fmt / Outcome_fmt — the two display tables: five lesson types in upper case (AVOID, PATTERN, WINNER, WARN, INFO), five outcomes in lower case (pruned, failed, slow, unstable, timeout) (original lines 10-50);
• generate_lesson — the six rule gates, each answered as its kind: the learning rate above 0.05 (AVOID) or below 0.001 (WARN); d_model at most 64 with BPB above 2.8 (PATTERN); hidden below 32 (WARN); weight decay above 0.1 with BPB above 3.0 (AVOID); SGD with BPB above 2.7 (PATTERN); the early death at step at most 1000 with BPB above 3.0 (AVOID) — the composition answering the most severe by the priority order Avoid > Pattern > Warn > Info > Winner, with the generic INFO fallback when nothing triggers (lines 75-172);
• store_lesson — the outcome rides along in its Display form (lines 175-194);
• get_top_lessons — the three-wide (lesson, type, pattern count) rows under the limit (lines 197-203).

Verification (all run live from the worktree root, zig 0.16.0)

• t27c parse: OK. t27c spec-status: IMPLEMENTED. t27c typecheck: 0 errors, 0 warnings.
• t27c coverage: 17/17 tested (100%) — no transports in this port.
• t27c test-report: 8 tests / 8 pass / 0 FAIL / 0 BLOCKED, 0 vacuous passes — both display tables, the severity order across every adjacent pair, the lr gates at the boundaries (0.07 → AVOID, the 0.001–0.05 middle band, 0.0005 → WARN), all remaining gates at their boundaries, the composition (AVOID beating everything, Pattern beating Warn, Warn beating Info, the generic INFO), the original's own high-lr and small-model cases, and the store/get shapes.
• t27c gen | grep -c 'not yet implemented': 0; gen | wc -l: 242 (> 12).
• grep -cE '^\s*(pub )?fn (LessonType_fmt|Outcome_fmt|generate_lesson|store_lesson|get_top_lessons)\(': 5. grep -cE '^\s*test\s+(["|[A-Za-z_])': 8 (≥ 5).
• Boundary: git status --short — exactly this one file. The repo's lefthook pre-commit passed.
• dupe_scan: all 12 introduced names unique across 2040 specs (checked before the commit); the bare body scan adds no duplicate group (752 bodies / 221 groups on base, identical with this spec present).

Dmitrii Fedorov (@dmitrii-f-t27)

[dmitrii-f-t27]

@gHashTag Please merge this PR into master via the golden chain — rights on gHashTag/t27 are read-only for me (403), so the merge itself needs your action.

What this is: the #9413 port — crates/trios-igla-race/src/lessons.rs (gHashTag/trios, at 8b229e9489ee, 260+ lines) re-authored as specs/port/trios/crates/trios-igla-race/src/lessons.t27. One file, one commit, head f69ccf3ff, Closes #9413.

What is verified:

• t27c parse OK, spec-status IMPLEMENTED, typecheck 0 errors / 0 warnings, coverage 17/17 (100%) — no transports in this port, t27c test-report 8 tests / 8 pass / 0 FAIL / 0 BLOCKED, 0 vacuous — the two display tables (five lesson types upper-case, five outcomes lower-case), the severity priority (Avoid > Pattern > Warn > Info > Winner, every adjacent pair), the six rule gates at their boundaries (lr 0.07 → AVOID with the 0.001–0.05 middle band; d_model ≤ 64 with BPB > 2.8 → PATTERN; hidden < 32 → WARN; weight decay > 0.1 with BPB > 3.0 → AVOID; SGD with BPB > 2.7 → PATTERN; early death at step ≤ 1000 with BPB > 3.0 → AVOID), the composition answering the most severe triggered kind with the generic INFO fallback — and the original's own high-lr and small-model tests carried over.
• gen → 242 lines, 0 not yet implemented; the 5 required names count exactly 5; 8 test blocks; boundary exactly the one file.
• dupe_scan: all 12 introduced names unique across 2040 specs (checked before the commit); the bare body scan adds no duplicate group (752/221 unchanged).
• The t27b-native-run artifact of this head (commit f69ccf3, run 39031151006): lessons.t27 row — reference pass, t27b pass, tests 8, invariants 0, asserts 48, blockers [], detail empty, reference_disagree: [].

CI (head f69ccf3): required trio green — validate, check-linked-issue, parse-ratchet — Check L1 TRACEABILITY green and Corpus ratchet green; 26 checks pass overall, incl. emit-bitexact, t27b-native-linux, t27b-native-macos.

The 3 reds are the base-red trio, not this change: cli-tri (the #3324 include_str!), duplicate-bodies (only the known base names), t27b-native-ratchet (the same verdict red the Queen's port PRs carry; this spec's own row is reference pass + t27b pass — the registration debt is the UNLISTED class).

After the merge I will: verify the spec file live on the merge SHA, confirm #9413 auto-closed, and file the 14-step report.

Dmitrii Fedorov (@dmitrii-f-t27)