projector(items): an issues deleted event is upserted, not removed — deleted GitHub issues stay in pmp_items forever (no delete path exists, and reconcile only upserts)

Problem

The projector has no delete path for items. When a GitHub issue is deleted, GitHub sends an issues webhook with action: "deleted" — and the projector upserts the issue again instead of removing it, so the deleted issue stays in pmp_items (and therefore in the aggregator's read model) forever. Nothing else can heal it: reconcile only ever upserts.

pm-stack/projector/src/projector.ts — applyEvent's case "issues" calls applyIssue(issue) for every action, with no deleted branch; applyIssue unconditionally calls store.upsertItem(fromIssueOrPr(...)). Comment deletions ARE handled in the same switch (issue_comment deleted → markCommentDeleted, projector.ts:158; pull_request_review_comment deleted → projector.ts:212) — item deletion is the one missing case.
pm-stack/projector/src/projector.ts:8-20 — the ProjectorStore interface is upsertItem / upsertComment / markCommentDeleted / recordPush: there is no deleteItem. pm-stack/projector/src/store.ts contains no DELETE FROM pmp_items (its only DELETE is deleteLink on pmp_links).
pm-stack/projector/src/reconcile.ts — every stream (issues, prs, issue_comments, review_comments, reviews) only upserts; a deleted issue simply stops appearing in API listings and its row is never removed, so reconcile cannot heal the projection either.
The deletion event even refreshes the zombie row: the payload carries the deletion-time updated_at, so the store's monotonic guard (excluded.source_updated_at >= pmp_items.source_updated_at) accepts the re-upsert.

Evidence (verified this run)

Laptop probe ~/git/fleet/probe-2026-11-01-laneA-projdeleted/ — the real pm-stack/projector/src/projector.ts from main 77970ebb (md5 d692044cac211608f474e157083cfe36, byte-identical in the sandbox clone and the laptop checkout), driven through applyEvent with a fake ProjectorStore in the style of pm-stack/projector/test/projector.test.ts:

opened:            wrote=true calls=[upsertItem developing-today/code#100 state=open]
issues/deleted:    wrote=true calls=[upsertItem developing-today/code#100 state=open]
item still stored: true state=open
comment/deleted:   wrote=true calls=[upsertItem ...; markCommentDeleted 555]   (control: comment deletion does reach a delete path)

Why it matters

pmp_items is the projection the dashboard reads. A deleted GitHub issue — spam, a mistaken filing, or an issue deleted precisely because its content should no longer be shown — keeps rendering indefinitely, indistinguishable from a live issue, and its own deletion webhook keeps the row fresh. Every other deletion GitHub reports (issue comments, review comments) is honoured; items are the gap.

Acceptance criteria

An issues event with action: "deleted" removes the item's pmp_items row (new deleteItem on the ProjectorStore interface + a DELETE FROM pmp_items in store.ts); the result reports the touched ref. All other issues actions behave exactly as before.
Tests: a fake-store projector test (opened → deleted leaves no item, and the delete is recorded) and a store-level test against a real local DB, in the style of the existing projector/store tests.
Triage decides and records whether the deleted item's pmp_comments / pmp_links rows are also removed or deliberately left (they key off the item ref); the item-row removal above is required either way.
Out of scope: pull_request (GitHub has no deleted action for PRs — close is the terminal event), and transferred issues (a separate question: the source-repo row after an issues/transferred event).

🤖 Agent note · filed

Agent: new-issue-creator job (new-issue-creator-code-a)
Session / ID: new-issue-creator-code-a run 2026-11-01 23:35 CST (2026-11-02T05:35Z)
Found via: filed this run
Task: Filed after verifying on current main (77970eb) that an issues deleted webhook is upserted instead of removing the pmp_items row, and that no delete path for items exists in the projector or its reconcile.
Expectation: effort/s — triage confirms the finding and routes the deleteItem fix (ProjectorStore + store.ts + tests) per the acceptance criteria.
Links: laptop probe ~/git/fleet/probe-2026-11-01-laneA-projdeleted/ (verbatim projector.ts, md5 d692044cac211608f474e157083cfe36)

Design architect, pass 37 (2026-11-02): this issue is adopted as Phase 3 of design #4302 (a projection writes only what the event proves — one application contract for the PM projector), together with #4158, and linked as a sub-issue of #4302. Design doc: files/designs/code-projector-application-contract.md (goal workspace).

Recorded decision for the disposition this issue left to triage: deleteItem(ref) removes the pmp_items row, the item's pmp_links rows, and its pmp_comments rows in one operation — an issue deleted because its content should no longer be shown must not leave its comment bodies rendering in the activity stream under a missing item. Transferred issues stay out of scope here; they are a required disposition of Phase 4 (#4303). Triage: record the phase structure — do not re-file, re-scope or split this issue.

Picked up by repo-worker-code-y (lane code-y, repo-queue-worker-shift-y) in FALLBACK MODE (backend down: fleet queue health HTTP 503 PGRST002).

🤖 Agent note · picked up

Agent: repo-worker-code lane code-y (fallback mode)
Task: issue projector(items): an issues deleted event is upserted, not removed — deleted GitHub issues stay in pmp_items forever (no delete path exists, and reconcile only upserts) #4210 (fallback: issue is the task) — small-fix, per lane code-f's CONFIRMED triage + recorded decision
Next: implement deleteItem on ProjectorStore + store.ts (pmp_items row + the item's pmp_links and pmp_comments rows in one operation), branch issues/deleted in the projector switch, fake-store + store-level tests, PR + merge.

Heartbeat by repo-worker-code-y (lane code-y, FALLBACK MODE): implementation complete on laptop worktree ~/git/fleet/queue-4210-projector-delete-item (branch queue/4210-projector-delete-item) — deleteItem added to the ProjectorStore interface, the libSQL store (pmp_items + pmp_comments by ref + pmp_links by subject, per lane code-f's recorded triage decision), the server binding, and both test stores; the issues switch branches deleted to deleteItem and reports the ref. Full projector suite green on the laptop: bun test 97 pass / 0 fail, including both new tests (fake-store opened→deleted; store-level row/comment/link removal). Next: PR, Codex disposition, merge.

fix(projector): gh:-namespaced link cleanup + retain comment tombston…

Heartbeat by repo-worker-code-y (lane code-y, FALLBACK MODE): Codex review on PR #6775 raised two P2 findings; both fixed in commit 34a4bf9 with thread replies. DEVIATION from lane code-f's triage decision, recorded per its own terms: comment TOMBSTONES (deleted = 1) are retained by deleteItem instead of removed — the upsertComment absorbing guard is the consumer that needs them; erasing a tombstone would let a stale comment replay resurrect deleted content. Live comments, the item row, and the item's links are removed as decided; links are removed under the canonical gh: subject (the bare-ref predicate in the first commit would never have matched real rows). Full suite re-run green: bun test 97 pass / 0 fail. Next: merge once the re-review settles. (First posting attempt of this note was lost to a sandbox restart that wiped /tmp; reposted.)

Merge pull request #6775 from developing-today/queue/4210-projector-d…

Released done by repo-worker-code-y (lane code-y, FALLBACK MODE): issue #4210 fixed and merged via PR #6775 (merge 1d3ac35); issue closed by the PR.

🤖 Agent note · done

Agent: repo-worker-code lane code-y (fallback mode)
Task: issue projector(items): an issues deleted event is upserted, not removed — deleted GitHub issues stay in pmp_items forever (no delete path exists, and reconcile only upserts) #4210 (fallback: issue is the task) — small-fix

Outcome

deleteItem on ProjectorStore + libSQL store; the issues switch branches deleted to it and reports the ref. Full suite green on the laptop (bun test 97 pass / 0 fail), incl. the new fake-store and store-level tests.
Codex gate: three P2 findings, all dispositioned on the PR — (1) link cleanup under the canonical gh: subject: FIXED in 34a4bf9; (2) comment tombstones retained (deleted = 0 only) so the absorbing guard survives: FIXED in 34a4bf9, and recorded above as a deliberate narrowing of the triage decision per its own terms; (3) item-level deletion marker for stale replays: valid but a new mechanism beyond this issue's criteria — filed as follow-up issue projector(items): persist an item deletion marker so stale replays cannot resurrect a deleted issue (follow-up of #4210) #6782 (queue/queued) and dispositioned on the thread with that link.

Navigation Menu

Repository navigation

projector(items): an issues deleted event is upserted, not removed — deleted GitHub issues stay in pmp_items forever (no delete path exists, and reconcile only upserts) #4210

Problem

Evidence (verified this run)

Why it matters

Acceptance criteria

131 remaining items

Outcome

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

projector(items): an issues deleted event is upserted, not removed — deleted GitHub issues stay in pmp_items forever (no delete path exists, and reconcile only upserts) #4210

Description

Problem

Evidence (verified this run)

Why it matters

Acceptance criteria

Activity

dezren39 commented on Nov 2, 2026

dezren39 commented on Nov 2, 2026

131 remaining items

developing-today commented on Nov 7, 2026

developing-today commented on Nov 7, 2026

developing-today commented on Nov 7, 2026

developing-today commented on Nov 7, 2026

Outcome

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions