Show a review's findings as a panel, not only in the agent's own words #288

Closed
opened 2026-10-28 19:24:24 -04:00 by cmoriarty · 1 comment
Owner

What happens

A review's findings can only be read in the agent's transcript, in whatever words it chose. Run 136's combined reviewer left a good summary; another run may not. The chip (#287) gives counts, not the findings.

Fix

A findings panel built from the review's record (the digest, so duplicates are merged and each finding names the lenses that raised it), not from the agent's prose:

  • A summary strip: 0 blocking · 6 to fix · 5 notes, with a bar split by the same colours.
  • Groups by weight: blocking (red), to fix (amber), notes (grey). Each finding is one row with a short title, the document (spec.md), the lenses (spec, test), the quoted line in mono, and what became of it (revised, or kept, because…).
  • A row opens to show the full problem, the suggestion, and a link to the document at the quote.
  • It shows under the review step's transcript and on the revise step, opens from the spine chip, and appears on the proposal gate. The code review's findings get the same panel.

Needed:

  • The reviewers write a short title per finding (≤ 60 characters); today problem is a paragraph.
  • The revision writes each finding's outcome as data, not only prose in revision-1.md.
  • The chip counts the digest, not the raw lens files. Run 136's chip said 7 to fix where the digest had 6, because the payoff-date finding came from two lenses.

Done when

From the spine, a person sees each finding of a review in one place, what it is about, and whether it was acted on, without opening a transcript.

From #279, run 136. With #287.

**What happens** A review's findings can only be read in the agent's transcript, in whatever words it chose. Run 136's combined reviewer left a good summary; another run may not. The chip (#287) gives counts, not the findings. **Fix** A findings panel built from the review's record (the digest, so duplicates are merged and each finding names the lenses that raised it), not from the agent's prose: - A summary strip: `0 blocking · 6 to fix · 5 notes`, with a bar split by the same colours. - Groups by weight: blocking (red), to fix (amber), notes (grey). Each finding is one row with a short title, the document (`spec.md`), the lenses (`spec`, `test`), the quoted line in mono, and what became of it (`revised`, or `kept, because…`). - A row opens to show the full problem, the suggestion, and a link to the document at the quote. - It shows under the review step's transcript and on the revise step, opens from the spine chip, and appears on the proposal gate. The code review's findings get the same panel. Needed: - The reviewers write a short `title` per finding (≤ 60 characters); today `problem` is a paragraph. - The revision writes each finding's outcome as data, not only prose in `revision-1.md`. - The chip counts the digest, not the raw lens files. Run 136's chip said 7 to fix where the digest had 6, because the payoff-date finding came from two lenses. **Done when** From the spine, a person sees each finding of a review in one place, what it is about, and whether it was acted on, without opening a transcript. _From #279, run 136. With #287._
Author
Owner

Shipped in PR #297 (1d98b25). Lenses give each finding a title; the revision writes .osf/review/r<n>/outcomes.json. GET /api/runs/{id}/review/{round} serves a round, and a panel under the review and revision steps lists it by weight: title, document, lenses, revised or kept, and opens to the quote, problem and suggestion. A click on a review chip opens it.

Verified on run 137: a click on the chip selected the step and opened the panel, 0 blocking · 6 to fix · 6 notes, every finding with the reviewer's own title ("Currency output varies with browser locale"). The revision wrote outcomes.json, all six to-fix findings revised with a sentence each, and GET /api/runs/{id}/review/1 serves them.

Shipped in PR #297 (1d98b25). Lenses give each finding a title; the revision writes `.osf/review/r<n>/outcomes.json`. `GET /api/runs/{id}/review/{round}` serves a round, and a panel under the review and revision steps lists it by weight: title, document, lenses, revised or kept, and opens to the quote, problem and suggestion. A click on a review chip opens it. Verified on run 137: a click on the chip selected the step and opened the panel, `0 blocking · 6 to fix · 6 notes`, every finding with the reviewer's own title ("Currency output varies with browser locale"). The revision wrote `outcomes.json`, all six to-fix findings `revised` with a sentence each, and `GET /api/runs/{id}/review/1` serves them.
Sign in to join this conversation.
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
cmoriarty/braid#288
No description provided.