fix(rules): document --since-on-squash-merge failure in refresh-world-model rule

[AceHack]

Substantive rule extension capturing today's 15:14Z empirical observation: `git log origin/main --since="N min ago"` is the wrong query for recent merges.

Empirical anchor

PR #3894 merged at 15:11Z. At 15:14Z:

• `git log origin/main --since="20 min ago"` → EMPTY (PR invisible)
• `git log origin/main --since="30 min ago"` → EMPTY (PR invisible)
• `git log origin/main | head -3` → shows backlog(B-0558): worktree-pool primitive — re-land of #3817 (BACKLOG.md conflict bypassed) #3894 at top ✓

Root cause

Squash-merge collapses N commits authored hours/days ago into one commit. The squashed commit's committer-date matches the merge time, but its author-date (the field `--since` filters on) is preserved from the original authoring time. PRs squashed from older branches drop out of recent `--since` windows even though they just landed on main.

Mitigation

Prefer `git log origin/main | head -N` for recent-merge queries. The `head` form sorts by topological / log order which IS merge-recency-correct. If a date filter is required, use `--committer-date-order` paired with explicit committer-date filtering.

Composes with

The local-ref-staleness pattern (`unable to update local ref` wedge) documented elsewhere in the rule. Both fail in the same direction — make recent activity invisible to naive queries. Verification: `git ls-remote origin main` returns ground-truth SHA regardless of local ref state.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 39dd02f394

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Correct incorrect git log --since semantics

This section documents a false root cause: it says --since uses author date, but git rev-list/git log commit limiting is based on commit timestamp, and --author-date-order vs --date-order only changes ordering, not which commits pass the filter. In practice, a commit with an old author date but recent committer date is still returned by --since, so this guidance can mislead triage by blaming squash merges instead of the actual stale-ref/fetch issue and cause operators to distrust valid date-window queries.

Useful? React with 👍 / 👎.

[Copilot]

Pull request overview

This PR updates the refresh-world-model-poll-pr-gate rule to document an observed failure mode where git log origin/main --since="N min ago" can miss recently squash-merged PRs, and to recommend safer “recent activity” queries during world-model refresh.

Changes:

• Adds a new rule section explaining why git log --since=... can hide recent squash merges.
• Recommends using git log origin/main | head -N (and related verification cues) for recent-merge checks.

Comments suppressed due to low confidence (2)

.claude/rules/refresh-world-model-poll-pr-gate.md:43

• P1: The explanation mixes terms in a way that’s internally inconsistent: it says --since filters on “commit date” and then later says --since uses author-date. In Git, “commit date” is often interpreted as the committer timestamp (CommitDate), so this wording can mislead readers. Suggest rewording to consistently use “author date” vs “committer date” (and optionally point to git log --pretty=fuller fields) to make the claim unambiguous.

Empirical anchor 2026-05-16T15:14Z: a squash-merged PR (e.g., #3894 merged at 15:11Z) was invisible to `git log origin/main --since="20 min ago"` even though `git log origin/main | head -3` correctly showed it at the top. Root cause: `--since` filters on **commit date** (preserved from original authoring), not **merge date**. A squash-merge collapses N commits authored hours/days ago into one commit whose `committer-date` may match the merge but whose `author-date` (used by `--since`) is the original authoring time. PRs squashed from older branches drop out of "recent" `--since` windows even though they just landed.

.claude/rules/refresh-world-model-poll-pr-gate.md:44

• P1: The mitigation suggests using --committer-date-order “paired with explicit author/committer date selection,” but --committer-date-order only affects ordering, not which timestamp --since/--until filters on. Please clarify what exact approach should be used when a time window is required (e.g., whether to filter by committer date explicitly and how), or drop this sentence to avoid prescribing a non-working/unclear command pattern.

**Mitigation**: prefer `git log origin/main | head -N` over `git log origin/main --since="N min ago"` when checking recent merges. The `head` form sorts by topological / log order which IS merge-recency-correct. If a date filter is required, use `--committer-date-order` paired with explicit author/committer date selection.