VS Code Sweeper — Adoption

Of the 1 acted-on closes, 1 were likely fixed instead — closed as completed without the sweeper's comment on a proposal other than implemented-on-main. Most plausibly the team fixed the issue in the meantime; these are not claimed as adoption (blue in the acceptance chart below).

2nd-reviewed before the close: 0 confirmed · 0 disputed · 1 never 2nd-reviewed — the verify lanes are newer than most of these closes; the split fills in from here. Details in Independent verification.

Proposed-close issues closed per week

Only closures of issues the sweeper proposed to close. Weeks start Monday.

Acceptance by close reason

Each bar is one close reason — the same merged rows as the verdicts page's verdict distribution (structured reason first, legacy close labels folded in; the label-free close is implemented on main). Right label: acted-on / proposals. "Closed otherwise" counts closures without the sweeper's comment — correlation, not proven adoption; blue splits out the likely-fixed-instead closes, which the sweeper doesn't claim at all. Newer proposals have had less time to be acted on, so compare reasons with similar proposal ages cautiously.

Comment adoption on the 1 acted-on proposals

1 of the 1 own-wording closes were likely fixed instead (closed as completed) — those proposed comments weren't passed over for a better one; the close had a different cause entirely.

sweeper-fix skill adoption

Draft PRs opened with the sweeper-fix skill, discovered by reconcile's seeded-by-marker search — per user, against their agent-fixable pool (the track). Corpus-wide pool: 6 reviewed fix specs. As current as the last reconcile run.

No sweeper-fix skill PRs discovered yet — the funnel fills in as maintainers use the skill (stamps land on each reconcile run).

Independent verification — the 2nd review

A blinded second review re-judges duplicate and implemented-on-main close proposals before anyone acts (the dashboard's "2nd review" badge). "Disputed" is a dispute to read, not a proven error — the blinded judge sees less context than the original review. Coverage counts open proposals (the verify sweep's selection universe); stale stamps (the record was re-reviewed since verification) are excluded from the mix, and live issue drift is re-checked by the nightly verify sweep, not this page. Scope is these two close reasons only.

duplicate: 4 of 4 open proposals verified · 1 closed unverified — implemented-on-main: 3 of 3 open proposals verified

Verdict mix by kind

Duplicates by the canonical's kind, implemented-on-main by claim kind (hover a row label for the lane). Right label: disputed / verified — a kind disputing well above its peers is a calibration signal for the review prompt.

The 4 disputed proposals are listed on the dashboard (the link lands pre-filtered); each ✗ badge carries the verifier's evidence. A dispute is worth a manual look before acting: either the review or the verifier is wrong, and both readings are on record.

Verified proposals by eventual outcome

Counts, not rates — closed-after-verification volume is still small, so percentages would mislead. A disputed proposal closed anyway and later reopened is a false positive the verify lane flagged in advance.

Closed despite a dispute

Closes that went ahead after the 2nd review refuted the proposal — either the human knew better (the dispute is advisory) or a false positive slipped past the warning; both readings are on record. The before-the-fact twin of the reversals queue below.

None yet.

Reversals — reopened after a close

Close proposals whose closure was later reversed (the issue reopened) — none so far. A reversal of a sweeper-comment close is the strongest false-positive signal the pipeline produces — worth a manual look (and a feedback issue) each. A reopen isn't automatically a wrong verdict: issues also reopen to track follow-up work or get re-closed differently.

Who acted on proposals

Attribution: poster of the matched comment, else who closed the issue.

All closures since review — context

Every reviewed issue that closed, by what the review had concluded. Only the propose-close row is claimed by the funnel above; needs-info issues auto-close on their own cadence, and keep-open closures happened for unrelated reasons (worth eyeballing as a precision check).