tri gates mutate: refute the equivalence claims instead of quoting them (Refs #2161)

[gHashTag]

# mutant-equivalent: <why> marks a survivor as unkillable by construction, and tri gates mutate printed it back as "claims equivalent: …".

The word claims was carrying real weight: nothing had ever checked one. Six sit in tools/.

A claim is a statement about the code. It ages with the code. And the run best placed to notice it has gone stale is the mutation run itself — it already built the mutant and already knows the verdict. The missing step was one comparison.

Measured against all six that already existed: none contradicted. That is worth exactly as much as the run that could have refuted them and did not, and nothing more.

Counts, not membership

A line can hold more than one mutable site — if a < 1 or b < 1: holds two — so a claimed line with one of two dead has been contradicted once. Keying on "did this line leave the survivor list" would call that claim intact, and a half-true claim is the hardest kind to catch by eye.

Writing about the marker created a claim

The extractor matched mutant-equivalent: anywhere inside any comment. So this sentence, written in this very PR —

that reasoning now sits on the line as a # mutant-equivalent: claim

— registered as a claim of its own, bound to whatever code line happened to follow.

Caught because the count printed 2 where I had written 1. Nothing else would have caught it: a claim nobody made, on a line it says nothing about, silently waiting to be reported as contradicted the day that unrelated line's mutant died. A false positive in the extractor becomes a false refutation in the checker — the new check gave the old bug a way to lie.

The marker must now open the comment. Every real claim in the tree already did.

The general shape: a scanner that matches a marker anywhere cannot tell a use from a mention, and documentation is exactly where mentions live. Any tool that greps the tree for its own vocabulary will eventually read its own documentation as data — and the count is the cheapest place to notice, which is an argument for printing counts nobody asked for.

The positive control, and how it nearly did not run

Planting a false claim over a line whose mutant is known to die must produce a contradiction. My first attempt printed nothing, and I read that as "the check does not fire".

It fired at nothing. Planting made tools/ dirty, mutate refuses to start on a dirty tree, and I was reading its output through grep -i CONTRADICTED — a filter that cannot show a refusal. The verdict and the error went to the same stream and I had closed the half that was speaking.

Satisfying the guard rather than bypassing it, it fired exactly once, naming the line and the claim.

Tests

contradicted_claims() is separate from the run loop — a checker nobody can test is the same failure one level up. Six cases: survived, died, half-died, claimed-but-not-mutated-in-this-direction, mention-is-not-a-claim, marker-opens-the-comment.

First claim written under the check: gft_backprop_microcode.py's if d >= 26, proven equivalent over 525,918 points in #2653.

Refs #2161

[github-actions]

PR Dashboard

Generated at: 2026-08-24 03:28:06 UTC

• tri gates mutate: refute the equivalence claims instead of quoting them (Refs #2161) #2656 tri gates mutate: refute the equivalence claims instead of quoting them (Refs #2161) | Created: 2026-08-24 | Updated: 2026-08-24 | 20 failing / 24 total
• Unbreak FPGA CI and correct two unbacked hardware claims #2081 Unbreak FPGA CI and correct two unbacked hardware claims | Created: 2026-08-11 | Updated: 2026-08-20 | 5 failing / 27 total

Summary

Status	Count
Total Open PRs	2
PRs with Failing Checks	1
PRs with All Checks Green	1
READY	0
FAILING	1
PENDING	0

Seal Status

• ⚠️ STALE -- sha256(compiler.rs)=e6333575ec10 != manifest seal=87e5cbd3ad94.
  The committed NMSE numbers were certified against an older compiler.rs.
  Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

[github-actions]

📓 NotebookLM Notebook linked to this PR

• Notebook ID: b83263109fb055dc
• URL: https://notebooklm.google.com/notebook/b83263109fb055dc

This notebook contains session context, decisions, and artifacts for this work.