fix(decode): fail-closed false-singleton + consume min_terminals bound

[Tyler-R-Kendrick]

Summary

• Fail closed when UNKNOWN witness drops leave a lone proven candidate (status=incomplete, reason witness_false_singleton_risk) so I2 cannot force a budget-manufactured singleton.
• Consume certified StaticLalrAdapter.min_terminals in terminal_witness as a negative-direction prune only.

Test plan

• decode_stats false-singleton tests
• static LALR adapter consumer allowlist + bound soundness
• version bumps dsl.operators.registry v13, dsl.completion_kernel v21

[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: 50 minutes

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

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: 9d70e5c6-279f-4e3b-838b-05838d9aae73

📥 Commits

Reviewing files that changed from the base of the PR and between ec48e28 and 92804b6.

📒 Files selected for processing (8)

• docs/design/hybrid-unmask-hx5-20260804.md
• src/slm_training/dsl/grammar/fastpath/completion_artifact.py
• src/slm_training/dsl/grammar/fastpath/completion_kernel.py
• src/slm_training/dsl/grammar/fastpath/static_control_domain.py
• src/slm_training/dsl/pack.py
• src/slm_training/resources/versions.json
• tests/test_dsl/test_static_lalr_adapter.py
• tests/test_models/test_decode_stats.py

---

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.}

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
slm-training	Ready	Preview	Aug 6, 2026 1:13am

[Tyler-R-Kendrick]

Rubber-duck + adversarial closeout review (agent)

Rubber duck

Walked the diff for PR #1457 in isolation vs its base. Intent matches the PR summary; docs + version stamps land with the code; tests pin the new invariants.

Adversarial

• Decode legality / I2: false-singleton path fails closed (incomplete domain) rather than inventing legality — does not weaken constrained decode or singleton bypass correctness.
• LALR bound: only negative-direction prune (room < certified lower bound → UNSUPPORTED). Certification required before use; failure disables prune for the session.
• CI weights: validator fails closed on partial / super-wall files without irreducible remediation; packing still degrades gracefully if table absent.
• Thrash regimes: production residual slugs stay constrained; confirm/promote freeze thrash; climb no-op skip uses materialized knobs not sparse overlays.
• Size / gates: no capacity growth; no ship-gate weakening.
• Coverage honesty: remaining topology live-apply and live dsh5 preflight stay @pytest.mark.slow — stated deferral, not silent skip.

No blocking findings. Proceeding with CI + squash-merge bottom-up.