Add unreviewed-merge metric to the reviewer table - #2

What

Adds an Unreviewed column (with a Requested denominator) to the Reviewers table: how often a PR was merged while that reviewer's requested review was still outstanding.

│ Name          ┆        p50 ┆ ... ┆ Count ┆ Requested ┆ Unreviewed │
│ babakks       ┆  69h10m12s ┆ ... ┆   101 ┆        80 ┆   16 (20%) │
│ BagToad       ┆  23h55m41s ┆ ... ┆    61 ┆        73 ┆   23 (32%) │
│ williammartin ┆   1h18m47s ┆ ... ┆    22 ┆        37 ┆   16 (43%) │

Why

The existing turnaround percentiles only measure reviews that were actually submitted, so a review that was asked for and then overtaken by the merge is invisible — it silently drops out of the stats rather than showing up as a slow one. This makes that case countable.

How

The GraphQL query now pulls each PR's REVIEW_REQUESTED_EVENT and REVIEW_REQUEST_REMOVED_EVENT timeline items. review_requests replays those events chronologically per reviewer to determine whether the request was still open at merge time, and reviewer_stats counts it as unreviewed when the PR merged with the request outstanding and no review from them submitted by mergedAt.

Three judgment calls worth a reviewer's attention:

• Withdrawn requests don't count — a request removed before the merge isn't a missed review.
• Requests added after the merge don't count — GitHub permits this, and it isn't a case of merging out from under someone.
• Team and mannequin requests are skipped — they name no individual to attribute the miss to. Bot requests follow the existing --include-bots flag.

Side effect: reviewers who were requested but never reviewed now appear in the table, which previously was keyed only on submitted reviews. Their percentile columns render as -.

Verification

cargo build and cargo clippy are clean.

Ran against cli/cli over a two-month window and cross-checked every reviewer against an independent Python implementation over the same raw API data — all six matched exactly (babakks 80/16, BagToad 73/23, williammartin 37/16, tidy-dev 21/7, sergiou87 2/0, niik 1/1). That data exercised bot requests (76, correctly excluded by default and included under --include-bots), team requests (231, skipped), and removal events (4).

Hand-verified cli/cli#14019: niik was requested at 11:23:58 and it merged at 11:31:05 with no review — correctly flagged.

No live PR exercised the "request added after merge" branch, so that path is reasoned-through rather than observed.

Notes

• cargo fmt --check still reports one diff at src/main.rs:594. That is pre-existing on master (confirmed via git show master:src/main.rs) and was left alone rather than adding unrelated churn.
• Unrelated and not fixed here: the README's Authors example table is stale from an earlier change — it is missing the Size (p75)/Size (p99) columns the code emits today.

🤖 Generated with Claude Code