Skip to content

gunbc-private-direction - #11361

Closed
gunbai-bot[bot] wants to merge 16 commits into
mainfrom
pr11097
Closed

gunbai-bot[bot] wants to merge 16 commits into
mainfrom
pr11097

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Sep 14, 2026

Copy link
Copy Markdown
Contributor

Auto-opened by session-dashboard for session quick-ibex-468.
Pushing to pr11097 advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

  • Title describes the change (not the session id or branch).
  • PR body summarises what and why (replace the TODO below).
  • Tests run: name the command (e.g. npm test, cargo test) and the result.
  • If this closes a work item, the body contains a Closes #N directive.
  • No commits on this branch are surprises (no fork/cherry-pick I did not make).
  • No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

  • TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

Brian Searls and others added 16 commits September 12, 2026 14:00
…xpressible without treating ListingGone as a delete.

The arm writes DerivedField only, refuses while the listing is still projected or standing is not ListingGone, and never touches judgement or drops the row.

Co-authored-by: Cursor <cursoragent@cursor.com>
…ular scaffold.

Preconditions lived in the DerivedField arm, so a schema with no derived columns could converge on a refused request. classify_review_sheet already consumes the three-arg fold; a frontier whose trigger was "bind this function" did not name a later change.

Co-authored-by: Cursor <cursoragent@cursor.com>
…e refuses at the identity cell.

The door was only reachable from tests and a zero-erasure wrapper. Ensure is the operator path; empty erasures remain the no-erasure pass. Row-grained ErasureRefused now names the identity A1, matching located write-or-refuse.

Co-authored-by: Cursor <cursoragent@cursor.com>
…cells record an erasure.

A later projection that re-emits the same listing id would otherwise rewrite title and flip standing with no diagnostic. Gone rows that still carry derived text still restore; that is a scrape gap, not an erasure.

Co-authored-by: Cursor <cursoragent@cursor.com>
…g durability.

An authorized request with no derived columns is a located refuse, not Converged. Gone-plus-empty cells are not an erasure record; the caller must re-supply the erasure list each pass.

Co-authored-by: Cursor <cursoragent@cursor.com>
…an read.

Empty derived cells are not an erasure record. The authorized pass writes DerivedErasureMark, and a later pass that would restore title from the projection keeps those cells blank because it reads that mark.

Co-authored-by: Cursor <cursoragent@cursor.com>
Applied is a readable label; blank is the unmarked cell. Column lookup goes through first_column_of_kind so each ReviewColumnKind is not its own filter copy.

Co-authored-by: Cursor <cursoragent@cursor.com>
Kind selection is now a keep over ReviewColumnKind. Schema-missing vs header-missing are distinct ObservedKindCell arms, and the identity-split deferral is a Scaffold bind instead of a comment.

Co-authored-by: Cursor <cursoragent@cursor.com>
Kind membership is exhaustive match that yields a column or a refuse arm, so the Bool keep helpers (and the keep-parameter fold) are gone.

Co-authored-by: Cursor <cursoragent@cursor.com>
Operator names are filter then map again; listing identity is listing_id, not a no-op scaffold bind.

Co-authored-by: Cursor <cursoragent@cursor.com>
Erasures are a parameter of classify, not a second name. Column lookup shares first_column_where; take is exhaustive match, not a Bool kind predicate.

Co-authored-by: Cursor <cursoragent@cursor.com>
Authorize already matches ObservedKindCell for schema-absent vs header-missing; a ListingStandingRead? wrapper had no call site and collapsed those causes.

Co-authored-by: Cursor <cursoragent@cursor.com>
Conflict markers were left in append_cell_for_column; keep parent skip of operator columns and the door's blank DerivedErasureMark, and restore columns_by_name so classify/append still resolve.

Co-authored-by: Cursor <cursoragent@cursor.com>
@gunbai-bot

gunbai-bot Bot commented Sep 14, 2026

Copy link
Copy Markdown
Contributor Author

Closing: this is a duplicate of #11097 on the identical head 3fae4269b0, created by the dashboard auto-opening a PR from a local branch I carelessly named pr11097 and pushed. The real PR is #11097 on session/still-crab-505-derived-erasure.

This one is not merely redundant, it is harmful: two PRs on one commit duplicate the required check contexts, which is why #11097 was reading BLOCKED with doubled required-witnesses-build / required-witnesses-floor names while the commit itself was fine. Branch deleted with the close so it cannot reopen.

— sent from quick-ibex-468

@gunbai-bot gunbai-bot Bot closed this Sep 14, 2026
@gunbai-bot
gunbai-bot Bot deleted the pr11097 branch September 14, 2026 10:53
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