unic-dlc-pr-review reconcile counts withdrawn, kept-by-design and moved findings as fixed

What the walk measured

unic-dlc-pr-review's reconcile headline counts a finding as fixed when it was withdrawn, deliberately
kept, or moved somewhere else. Measured 2026-08-26 on
PR 5798 in DXP-DesignSystem, the
first two-iteration run of this Box against a real pull request.

Iteration 2 posted:

12 fixed · 1 still present · 0 regressed · 8 new

Four of that twelve were not fixed:

1. Kept by design — a passWithNoTests flag the author answered no-change, with a reason, because the
   prefactor ships no tests.
2. Kept by design — a tsconfig paths mapping serving an add-then-reshape window.
3. Moved, not closed — a column-width finding whose rewrap left a different line over the limit.
4. Withdrawn — a scope finding whose confidence dropped below the posting threshold on re-read.

The Box's own text is the evidence

The first entry under Fixed since iteration 1 reads:

ADR-0003 undercounts the prefix edit sites: three, not two — the ADR now says three. The undercount
survives on WI 42974 item 4, raised below as a new finding.

An entry in the fixed bucket stating that the problem survives elsewhere is the defect in one sentence.
The per-finding prose is honest throughout — every one of the four says what really happened, in its own
line. Only the tally is wrong.

And the Box contradicts itself inside one run. It set the two by-design threads to byDesign on the pull
request — thread 65283 and thread 65287, both verified — while counting them as fixed in the summary. The
thread status is right and the headline is wrong, from the same run.

Why it matters

The headline is what a human reads. A reviewer who sees 12 fixed · 1 still present believes one thing is
outstanding when four are: two permanently by choice, one moved, one withdrawn. The two by-design ones are
the dangerous pair, because a deliberate deferral counted as a fix has lost its expiry date — which is
exactly what this walk had to carry by hand into the next slice's acceptance criteria.

What a fix looks like

The buckets, not the arithmetic, are the problem. Four states rather than two:

• fixed — the finding is gone and the Box verified it.
• kept by design — the author answered no-change with a reason the Box accepted. Needs somewhere for the
  expiry to be named, or it becomes permanent silently.
• withdrawn — the Box no longer stands behind the finding, with the reason (confidence, re-read).
• moved — the same defect now lives somewhere else, naming where.

still present and regressed keep their current meaning. Whatever the buckets are called, a finding the
author deliberately kept must never be counted in the same number as one that was fixed.

Related

• #430 — the same run could not verify the author's
  build, lint, type and test claims. Different defect, same review.

Start in the unic-dlc-pr-review reconcile logic that produces the headline counts, using the iteration 2 results and the four examples in this issue as the behavioral reference. Trace how fixed, byDesign, withdrawn, and moved findings are bucketed and reported; done means deliberately kept, withdrawn, and moved findings are no longer included in the fixed count and their reasons or destinations remain visible.