Skip to content

fix(tri): --invert was a banner over a silent run; add --all (#2492) - #2500

Merged
gHashTag merged 3 commits into
masterfrom
feat/mutate-all
Aug 23, 2026
Merged

gHashTag merged 3 commits into
masterfrom
feat/mutate-all

Conversation

@gHashTag

Copy link
Copy Markdown
Collaborator

tri gates mutate --invert shipped in #2499, merged, and published a number. It never ran the invert operator.

Direction::Invert was declared and documented. invert_sites() was written and unit-tested. Nothing joined them: mutate() chose its direction with if loud { Loud } else { Silent }, so the flag printed an invert banner over a silent run.

Proven two ways before touching it

  • static: zero calls to invert_sites() in non-test code, Direction::Invert never constructed
  • empirical: --invert and the default printed byte-identical rows

Why it stood

The published result — "one survivor, check_elab_ratchet's no-baseline branch, the same one both other operators leave" — was true in every part, and the last clause was true by construction: it was the other operator. A real measurement of the wrong thing agrees with the story you already have.

All ten existing tests passed with the bug present. Each exercised invert_sites() or sites_in_direction() alone; none crossed between them — the defect this command exists to find, one level up in the auditor.

Changes

  • sites_in_direction delegates Direction::Invert to invert_sites
  • each_direction_reaches_its_own_operator: three distinct sites on one fixture, asserting the three site sets differ. Verified RED on the restored bug, green without it.
  • --all: three operators, one pass, three columns. Reading them apart is what let the substitution stand.
  • the NOW entry carrying the wrong number is retracted in place, with the real one

First honest measurement

33 invert mutants across 13 gates, 33 killed, no survivors. The columns disagree where they must: check_json_parses one silent site and no invertible condition, check_vector_data five inverts against four silents. The suite's one remaining survivor is unchanged — check_elab_ratchet.py's no-baseline branch under the silent operator, declared UNCOVERED.

Refs #2492

Direction::Invert was declared, documented and never constructed.
invert_sites() was written, unit-tested, and had zero callers. mutate()
chose `if loud { Loud } else { Silent }`, so `--invert` printed an
invert banner and measured the silent operator.

Proven two ways before the fix: statically (0 calls, variant never
constructed) and empirically (--invert and the default printed byte
identical rows).

The published "one invert survivor, the same declared branch" was the
silent survivor relabelled. It read as correct BECAUSE it was a real
measurement of the wrong thing -- number-audit 8.6: the observation was
right and the reason was invented.

All ten existing tests passed with the bug present; each exercised
invert_sites() or sites_in_direction() alone and none joined them. That
is the defect this command exists to find -- function covered, wiring to
the answer not -- arriving one level up, in the auditor.

Adds --all: three operators, one pass, three columns. Reading them apart
is what let the substitution stand.

Refs #2492
The published "one invert survivor" was the silent survivor relabelled.
Real first measurement: 33 invert mutants across 13 gates, 33 killed.

ci-gates SKILL.md §36: a flag can be a banner. The output was plausible
and "the same branch as the other operators" was true by construction --
it WAS the other operator. Check: assert a new mode produces something
the others do not, on a fixture where all have a site. Equality across
modes is the signature of a fall-through.

Refs #2492
@github-actions

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

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

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-08-23 04:49:54 UTC

Summary

Status Count
Total Open PRs 6
PRs with Failing Checks 2
PRs with All Checks Green 4
READY 3
FAILING 2
PENDING 0

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e905ba8eeb53 != 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).

@gHashTag
gHashTag merged commit 06d44e2 into master Aug 23, 2026
29 of 31 checks passed
@gHashTag
gHashTag deleted the feat/mutate-all branch August 23, 2026 04:59
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant