Build replay-verified solver supervision corpus (VSS3-01, SLM-69)

[Tyler-R-Kendrick]

VSS3-01 (SLM-69): replay-verified solver supervision corpus

Adds a torch-free harness that turns main's canonical solver replay event-stream into a supervised support-set corpus for VSS3 residual guidance. Replay verification is enabled by default — only replay-valid states and full-mode-verified certificates produce hard labels.

What's here (5 files, +1496/−3)

• src/slm_training/harnesses/distill/solver_supervision.py — SolverSupervisionBuilder replays each trace (replay_violations) before emitting any row, then partitions each live hole domain into supported / unsupported / unknown with per-value certificate_ids_by_value / witness_digests_by_value. UNSUPPORTED comes only from a certificate-backed certified_deduction; SUPPORTED only from a verifier-accepted witness; all supported alternatives are retained (the chosen value is not privileged). Cost-to-go is censored with explicit reasons on truncated / budget-stopped / nonterminal suffixes; a local nogood is a hard-negative feature, never a global relabel. Cross-split state-leak guard; only ids/kinds/digests are stored (no raw source text); deterministic hashing.
• scripts/build_solver_supervision.py — CLI producing manifest.json, rows/support_set.*.jsonl, and validation_report.json; --describe dry-run writes nothing.
• tests/test_harnesses/distill/test_solver_supervision.py — 14 tests covering the honesty invariants.
• src/slm_training/harnesses/distill/__init__.py — defer the torch-backed sft import via PEP 562 __getattr__ (mirrors models/__init__.py) so the torch-free module imports without torch; public API unchanged.
• docs/design/iter-vss3-01-solver-supervision-20260718.md — dated build note.

This PR was reduced to exactly the in-scope VSS3-01 change and rebased directly on main — the earlier branch's out-of-scope re-implementations of dsl/solver/, model code, dashboard assets, README/MODEL_CARD, and CI config were dropped (they conflict with or duplicate work already on main).

Honesty

No model or head is trained; nothing is published. Consuming this corpus to train a cost/energy head is the next issue (VSS3-02 / SLM-70). No ship gate is touched or weakened; fixture rows would establish wiring only and make no model-quality claim.

Verification (against current main)

• pytest tests/test_harnesses/distill/test_solver_supervision.py → 14 passed
• solver no-regression (test_solver_state / test_solver_closure / test_capsule_solver) → 62 passed
• ruff check → clean; python -m scripts.repo_policy → ok; git diff --check → clean
• Pre-existing/unrelated: test_solver_trace.py fails at collection (torch not installed) identically on untouched main.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
slm-training	Ready	Preview, Comment	Jul 18, 2026 6:15am

[coderabbitai]

Warning

Review limit reached

@Tyler-R-Kendrick, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 46 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: b8327e12-8ad5-445f-985d-324a332e92d1

📥 Commits

Reviewing files that changed from the base of the PR and between ef483a7 and ad67509.

📒 Files selected for processing (5)

• docs/design/iter-vss3-01-solver-supervision-20260718.md
• scripts/build_solver_supervision.py
• src/slm_training/harnesses/distill/__init__.py
• src/slm_training/harnesses/distill/solver_supervision.py
• tests/test_harnesses/distill/test_solver_supervision.py

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/slm-69-solver-supervision-corpus

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}