spec(port): port specs/port/trios/crates/trios-data/src/lib.t27 (Closes #9499)

[dmitrii-f-t27]

Summary

Port gHashTag/trios:crates/trios-data/src/lib.rs (at 8b229e9489ee, 123 lines) as specs/port/trios/crates/trios-data/src/lib.t27 — Roadmap Stage 6 (goal #4548), target t27c gen-rust. Closes #9499.

The original is the batched FineWeb data loader stub: new builds a FineWebBatch over a dataset path, next_batch answers batches of dummy token sequences until the stub epoch limit, offset / batch_size read the state, reset rewinds to the start.

Per the issue's rule ("Port the decision, not the plumbing"), the five listed functions are ported under their names as pure functions over the loader state — none moves data in or out, so none carries an undefined; transport body; the growable Vec<Vec<u32>> is the plumbing, and a FineWebBatchOutcome struct carries the batch decision instead (availability, sequence count, tokens per sequence, next offset):

• FineWebBatch_new — the initial state: offset 0, the 1000-sample stub epoch limit, the data path kept in the signature exactly as the original keeps its unused stub parameter (lines 24-30);
• FineWebBatch_next_batch — the core decision: exhausted once offset >= max_samples (None), otherwise batch_size sequences of 128 tokens advancing the offset by batch_size (lines 40-53);
• FineWebBatch_offset / FineWebBatch_batch_size / FineWebBatch_reset — the readers and the rewind to offset 0 with everything else kept (lines 56-68);
• sequence_length (128), dummy_token_id (42), stub_epoch_limit (1000) — the stub content constants (lines 28, 45-48).

Rust's Option becomes the availability verdict on the outcome; the &mut self advance becomes the next_offset the caller feeds back through the state.

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: 8/8 tested (100%) — no transports to leave untested.
• t27c test-report: 6 tests / 6 pass / 0 FAIL / 0 BLOCKED, 0 vacuous (20 runtime asserts), including the original's own three tests (test_new_batch_loader, test_next_batch_returns_correct_size, test_multiple_batches — the exhaustion loop's boundary pinned directly at 996 + 4 = 1000).
• t27c gen | grep -c 'not yet implemented': 0; gen | wc -l: 95 (> 12).
• grep -cE '^\s*(pub )?fn (FineWebBatch_new|FineWebBatch_next_batch|FineWebBatch_offset|FineWebBatch_batch_size|FineWebBatch_reset)\(': 5. grep -cE '^\s*test\s+(["|[A-Za-z_])': 6 (≥ 5).
• Boundary: git status --short — exactly this one file.
• python3 tools/dupe_scan.py --name <fn> for all 8 names on a clean origin/master worktree: no function named … in 2039 spec(s) — all unique.

Dmitrii Fedorov (@dmitrii-f-t27)

[dmitrii-f-t27]

Correction of my earlier comment on this PR. The earlier text asked for a merge and said the red checks were base failures only. That was wrong while CI is red, and it is withdrawn. This is not a new merge request.

Current state (head ae88001): 26 checks success, 3 failure. The CI is not green.

The port: specs/port/trios/crates/trios-data/src/lib.t27 only, Closes #9499. Its own tests: 6 of 6 pass, 20 runtime asserts, checked natively with t27b on the latest base and by the reference.

The three failures, precisely:

• cli-tri: cli/tri/src/gates.rs:9012 does include_str! of .github/workflows/brain-seal-refresh.yml. That workflow was deleted in ci: one queen-publish job opens the derived-record PRs the labs' verdicts allow (Closes #9092) #9117 (ci: repair the two workflows whose failure was in the workflow #3324 only removed its push step). It fails the same way on master. Tracked in cli-tri: retire deleted-workflow test and refresh measured census pins #9634; the fix needs an owner decision, see cli-tri: retire deleted-workflow test and refresh measured census pins #9634 (comment).
• duplicate-bodies: 12 violations on master (groups of the gen_w ports and the STEP1/STEP2 ports). The full master CI artifact was compared. That none of this spec's helper names appear in those groups is not the proof; the proof is the master run itself. gen_w part: spec(port): share tree helpers across eight gen_w ports (Closes #9633) #9636 (issue duplicate-bodies: share tree helpers across eight gen_w bench ports #9633); STEP1/STEP2: ci: remove copied helpers from STEP1/STEP2 t27 ports #9631.
• t27b-native-ratchet (run 38758247517): one native refusal on master (rust_convert, CSE hoist, ci: restore rust_convert native pass (reference CSE hoist regression) #9632) plus a backlog of unlisted specs. In addition this PR has one unlisted row of its own: specs/port/trios/crates/trios-data/src/lib.t27 passes natively but the ledger docs/reports/t27b_expectations.json has no row for it. A missing ledger row of this port is not a pure base failure; it needs the ledger refresh (t27b-native ratchet is red on every master run since it landed: ledger 099ac2224 predates 16 improvements and 40+ new specs #9600) or a one-line data row. The mesh-node crypto mismatch is mesh-node port: stop random keypair test from reading undefined primitive results #9635 (PR t27b: repair crypto and rust_convert; distinguish PR ledger lag #9637).

Tracking: #9631 (STEP1/STEP2), #9632 (rust_convert), #9633 with PR #9636 (gen_w), #9635 with PR #9637 (mesh-node crypto), #9634 (CLI), #9600 (ledger).

@gHashTag: please keep the golden-chain merge pending until every check on the last SHA is green. Nothing is requested now.

Dmitrii Fedorov (@dmitrii-f-t27)