blog: the scanner scored what it could not see

[gHashTag]

Ninth post of the gate-audit campaign, and the third in a row where the defect was in the instrument rather than in what it measured.

A mutation tool reported pack_index_consistency_gate.py as having no failure path to break. The gate has four. All are ternaries — return 0 if not fails else 1 — and the scanner matched only a bare return 1. A path it cannot see is a path it scores as covered.

Across twelve gate scripts: 34 seen, 8 missed.

The post's centre is not the fix. It is that repairing the instrument raised the survivor count from 13 to 20 — the second time in this campaign that mending a measurement made the number look worse and be truer (the first was an elaboration ratchet where removing syntax errors increased the error total, because a syntax error truncates the file and hides what is behind it). Both times the instinct is to distrust the repair.

It also names the asymmetry behind the deliberately conservative digit test: a missed site stays an open question, an invented one gets published as a defect in somebody else's work — which this tool had already done once, in an issue and a post, before anyone checked it.

And it lays the three defects side by side: first flag instead of the set, a 1:1 control map, one syntactic form. All the same class, all in the auditor.

en + ru. build:ci clean from apps/website, tsc clean on the blog data, typecheck ratchet reports no file gained errors.

Refs gHashTag/t27#2468, gHashTag/t27#2470