Skip to content

A later results review can supersede an earlier one, by name (1.15.0) - #49

Merged
pierg merged 2 commits into
mainfrom
review-supersedes
Nov 18, 2026
Merged

pierg merged 2 commits into
mainfrom
review-supersedes

Conversation

@pierg

@pierg pierg commented Nov 18, 2026 •

Copy link
Copy Markdown
Owner

Why

A reviewed row that has to be read another way, to the same result, could not be fixed. In proofs-lab, G4.Q2.15's shell-denied row read the run by a fixed path, which lab check refuses ("command never reads $RUN"). The row was rewritten to read $RUN, with the same value 0, and a second review passed it. The history check still said "a row that a review had passed in 1274c4d8 is gone or changed", and putting the old row back brought back the $RUN failure. The gate on proofs-lab's main has been red since.

What changes

  • lab review <experiment> results pass "<why>" --supersedes <of> names the earlier passing results review by its of (8 to 16 hex digits). It ends the note with "Supersedes the results review of .". Once that review is committed together with the rows it read, the history check holds those rows instead of the ones the earlier review read; until then it supersedes nothing.
  • A later review that does not name the earlier one still supersedes nothing, so the existing guarantee (refilled rows reviewed again stay held) is unchanged. The explicit flag was chosen over letting any later pass supersede, because that existing test asserts the opposite and the flag leaves a written reason.
  • lab review refuses --supersedes on a plan review or a failing review, for an of that no earlier passing results review has, and when the earlier review read the rows as they are now; it refuses a note that ends with the sentence typed by hand. lab check refuses a note that names no earlier passing results review of other rows.
  • docs/model.md (Reviewed rows stay, reviews.tsv format), the lab-review skill (step 5), CHANGELOG and version 1.15.0.

Tests

Three new tests in tests/test_held.py: the refusals and the supersession with the new rows held in their turn; a forged marker; and a superseding review committed without its rows, which frees nothing. The full suite passes locally except two tests about stopping a dead run's processes, which fail the same way on main in that container.

@pierg
pierg merged commit a617ea7 into main Nov 18, 2026
4 of 5 checks passed
@pierg
pierg deleted the review-supersedes branch November 18, 2026 17:27
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