# 0028. The phase review is capped and names its uncertainties

- **Status:** Accepted
- **Date:** 2026-08-08

## Context

`docs/WORKFLOW.md` §3 asked for a "very short" summary in three categories: what changed, what needs a human
eye, what could not be verified. In practice "very short" does not hold — an adjective is not a limit, and
the summary drifts back towards a progress report.

The three categories also share a blind spot. All of them are about *coverage*: what was touched, what is
left. None of them asks about *confidence*. The review's real value is catching a judgement call that went
the wrong way — an assumption about what was meant, a case the code now handles differently — and nothing in
the format invited that.

## Decision

**Ten lines at most, and a fourth category for what the work is least sure about.**

- **Ten lines.** A number, not an adjective. A category with nothing in it is left out, not written out as
  "nothing here".
- **What you are least sure about** — at most three, one line each, ranked. Not what could not be checked:
  what *was* checked and might still be wrong.
- The three conditional rules — a dormant state becoming reachable, a URL-structure change, a schema change —
  stay exactly as they were.

## Alternatives

- **Keeping "very short".** It had already been tried; the drift is the evidence.
- **A word count, or a fixed template with headings.** Both punish the small phase, which is most of them. A
  line cap scales down to two lines without ceremony.
- **Folding uncertainty into "what needs a human eye".** They read the same and are not: one is *I cannot
  judge this*, the other is *I judged it and might be wrong*. Merged, the second disappears — it is the one
  that feels like admitting fault, so it needs its own slot to survive.

## Consequences

- The uncertainty list is only useful while it stays honest. Three confident lines of "nothing much" is worse
  than no category, because it reads as a clean bill of health.
- The cap forces a choice about what matters in a large phase. That is the intent — a phase whose review
  cannot fit in ten lines is usually a phase that should have been two.