Add Stack Integration + Legal TTT submission package

[Taleef7]

Summary

• add the 2026-03-28 Stack Integration + Legal TTT + Parallel Muon submission folder
• include README.md, submission.json, train.log, and the three audited train_seed*.log evidence files
• keep the PR diff limited to records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/

Verification

• python experiments/audit_submission_package.py records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/train_seed1337.log records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/train_seed42.log records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/train_seed2025.log
• reviewed clean-worktree diff scope against origin/main before push

[Copilot]

Pull request overview

Adds a new Track (10min / 16MB) submission package under records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/, bundling the promoted training script plus audited 3-seed evidence logs and metadata for reviewer consumption.

Changes:

• Add promoted train_gpt.py submission script and canonical train.log plus 3 audited train_seed*.log evidence logs.
• Add README.md describing provenance, run contract, and evidence inventory.
• Add submission.json capturing audited metrics and artifact size metadata.

Reviewed changes

Copilot reviewed 3 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/README.md	Reviewer-facing provenance + run/verification instructions for the promoted submission package.
records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/submission.json	Audit-derived metadata (metrics, sizes, seeds) for the submission package.
records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/train_gpt.py	The promoted training/eval script included as the submission’s executable artifact.
records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/train.log	Canonical log (intended alias of seed 2025) for quick review.
records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/train_seed42.log	Seeded evidence log (42) for the audited 3-seed package.
records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/train_seed2025.log	Seeded evidence log (2025) used as the canonical train.log content.
records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/train_seed1337.log	Seeded evidence log (1337) for the audited 3-seed package.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

README references experiments/verify_run.py, but there is no experiments/ directory in this repo. This makes the documented run/verification command unusable; please update the instructions to point at an existing verifier in the repository, or include the verifier script in this submission folder.

[Copilot]

README suggests regenerating metadata via python experiments/audit_submission_package.py, but that script/path does not exist in this repository. Either update the documentation to the correct script location/name or add the referenced audit helper to keep the package reproducible.

[Copilot]

bytes_code appears inconsistent with the evidence logs: in train_seed2025.log, Total submission size int6+lzma (15,990,006) minus Serialized model int6+lzma (15,900,548) implies a code size of 89,458 bytes, but this JSON records 91,356. Please correct bytes_code (and any derived fields) to match the audited artifact size computation.

Suggested change

	"bytes_code": 91356,
	"bytes_code": 89458,

[MatoTeziTanka]

Community Review — Add Stack Integration + Legal TTT submission package

BPB: (not parsed — see PR title) | Compliance: LOOKS CLEAN — score-first-per-chunk TTT (legal #1416/#1423 pattern)

What I found in the code (head SHA ef4d3f3ae3ee, file records/track_10min_16mb/2026-03-28_StackIntegration_LegalTTT_ParallelMuon/train_gpt.py):

The TTT path at line 1074 implements the score-first-per-chunk pattern: each chunk is scored under torch.no_grad() / inference_mode() before the base_model.train() + SGD adaptation runs on that same chunk, with an is_last_chunk guard so the final chunk gets no adaptation pass. This is the structural shape the legal frontier uses (PRs #1416 erichroepke, #1423 aryanbhosale).

Per Issue #402 and Issue #677, TTT is legal when each token is scored before the adapter updates on it, and that's what the code does here — chunk ci is scored under weights adapted only on chunks 0..ci-1. No prequant_ttt_adapt_adamw(val_tokens, ...) multi-epoch fine-tune, no scored-region SLOT, no target-in-key n-gram cache.

CPU smoke test (CT2038 proteus-engine, 2026-04-11): import OK in 0.05s, dim=512, layers=11, vocab=1024, code=89458 B, SMOKE_TEST_PASS

Verdict: LOOKS CLEAN.

Recommendation to @cocohearts @valerio-oai @0hq @yuzhougu-oai @notapplica: MERGE pending standard checks (3-seed validation, 16MB artifact cap, 10-min wallclock on 8×H100 SXM). The compliance picture matches the legal reference frontier and no flags were raised by the classification pass.

Auto-classification caveat: this review was drafted by the AST-based classifier against a template derived from manually-reviewed cluster PRs (#1420, #1450, #1487, #1541, #1529, #1533, #1518). If I've misread a subtlety in your eval path — e.g., multi-epoch TTT that I mistook for single-pass, or a target-in-key lookup I missed in a helper function — please flag it and I'll re-run the audit manually.

---

Reviewed by @MatoTeziTanka — The Agora. CPU smoke test (CT2038 proteus-engine, 2026-04-11): import OK in 0.05s, dim=512, layers=11, vocab=1024, code=89458 B, SMOKE_TEST_PASS. Classification via deterministic AST-based classify_prs.py (pattern bank derived from ~65 manually-reviewed PRs earlier in the 2026-04-11 sweep). This review was auto-drafted from a template and spot-checked before posting — if the template misread your code, please call it out so I can iterate the classifier.