Skip to content

fix(tri): the loud operator excluded ternaries on a reason I invented (#2492) - #2496

Merged
gHashTag merged 2 commits into
masterfrom
fix/loud-takes-ternaries
Aug 23, 2026
Merged

gHashTag merged 2 commits into
masterfrom
fix/loud-takes-ternaries

Conversation

@gHashTag

Copy link
Copy Markdown
Collaborator

Yesterday two gates reported no success path to break and I documented that as an honest limit, with a reason:

A ternary can yield 0 on one arm and a verdict on the other; forcing the whole line to 1 is the Silent operator's job seen backwards.

It is not. Silent forces the line to 0 — the gate never fails. Loud forces it to 1 — the gate always fails. Two different mutants of one line.

Excluding one printed a clean-looking row for a gate whose every verdict is a ternary. That row was an absence of measurement.

Allowing ternaries found two more survivors, and both were false

first_line()      return out.splitlines()[0][:88] if out else "(nothing)"
_is_exact_zero()  return v == 0

Helper functions returning values. Both predicates scanned for a standalone digit anywhere in the expression, so an index and a comparison read as verdicts. The silent predicate had the same weakness for 1..4: return v == 1 would have been a site.

Strict now: a return is a verdict when the whole expression is a literal, or a ternary whose two arms are literals.

The denominators moved downward

gate sites before after
check_elab_ratchet 10 6
wp18_conformance_gate 6 2
check_catalog_count 4 3
check_duplicate_agreement 4 3

Those earlier kills were real — the controls did notice. They were noticing mutations of helper functions. A higher denominator looked like more thorough measurement and was measurement of the wrong thing.

Where both operators stand

  • Loud: all killed everywhere except wp18_conformance_gate, whose success return is return code — a variable, correctly not a site, and the row now says no success path to break for a reason that is true.
  • Silent: one survivor, the branch declared UNCOVERED with its reason.

Three of my justifications have been wrong in this campaign

Two produced missing coverage. This one produced a row that read as a score — which is worse, because the first kind gets found by the next person who looks and the second gets quoted.

Refs #2492

…#2492)

The comment said: "forcing a ternary's whole line to 1 is the Silent
operator's job seen backwards". It is not. Silent forces the line to 0 --
the gate never fails. Loud forces it to 1 -- the gate always fails. Two
different mutants of one line.

Excluding one printed `no success path to break` for a gate whose every
verdict is a ternary. That row read as a clean score and was an absence
of measurement.

Allowing ternaries found two more survivors immediately, and both were
FALSE: first_line() returning a string with an index in it, and
_is_exact_zero() returning `v == 0`. Both predicates scanned for a
standalone digit anywhere in the expression, so an index and a comparison
read as verdicts. The silent predicate had the same weakness for 1..4:
`return v == 1` would have been a site.

Strict now: a return is a verdict when the whole expression is a literal,
or a ternary whose two arms are literals.

Denominators moved downward -- elab_ratchet 10 -> 6, wp18 6 -> 2,
catalog_count 4 -> 3, duplicate_agreement 4 -> 3. Those kills were real
and were noticing mutations of helper functions. A higher denominator
looked like more thorough measurement and was measurement of the wrong
thing.

Loud is now all-killed everywhere except wp18_conformance_gate, whose
success return is a variable and correctly not a site.

Refs #2492
Two gates reported "no success path to break" on a reason I wrote:
"forcing a ternary's whole line to 1 is the Silent operator seen
backwards". It is not -- silent forces 0, loud forces 1, two mutants of
one line. The sentence had the shape of a careful distinction, which is
why it lasted, and it printed a row that read as a clean score while
being an absence of measurement.

Plus: a predicate that scans for a digit is not a predicate for a
verdict. Allowing ternaries found two survivors and both were false --
an index and a comparison read as verdicts. Strict: a literal, or a
ternary of two literals.

And: a denominator that grows can mean worse measurement. The counts fell
when the predicate tightened; the earlier kills were real and were
noticing mutations of helper functions.

Refs #2492
@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-08-23 03:43:14 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).

@github-actions

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

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

@gHashTag
gHashTag merged commit 5ed45d1 into master Aug 23, 2026
31 of 33 checks passed
@gHashTag
gHashTag deleted the fix/loud-takes-ternaries branch August 23, 2026 03:50
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