feat(github): hold sibling PR checks and comment when an apply preflights a target

pkg/github/client.go

@@ -637,6 +637,37 @@ func (ic *InstallationClient) CreateIssueComment(ctx context.Context, repo strin
 	return comment.GetID(), comment.GetNodeID(), nil
 }
 
+// HasIssueCommentWithMarker reports whether any of the PR's most recent
+// comments contains the given marker string. Callers use it to make
+// at-least-once comment posting idempotent: a hidden HTML marker in the body
+// identifies the comment, and a retry that finds the marker skips the
+// re-post. Only the newest page of comments is inspected — the markers
+// callers search for are posted moments before the search, so an older
+// occurrence beyond the page means a duplicate is possible but harmless.
+func (ic *InstallationClient) HasIssueCommentWithMarker(ctx context.Context, repo string, pr int, marker string) (bool, error) {
+	owner, repoName := splitRepo(repo)
+	comments, err := retryGitHubUnavailableRead(ctx, ic.logger, "list issue comments", []any{"repo", repo, "pr", pr}, func(ctx context.Context) ([]*gh.IssueComment, error) {
+		list, _, callErr := ic.client.Issues.ListComments(ctx, owner, repoName, pr, &gh.IssueListCommentsOptions{
+			Sort:        new("created"),
+			Direction:   new("desc"),
+			ListOptions: gh.ListOptions{PerPage: 100},
+		})
+		if callErr != nil {
+			return nil, classifyGitHubAPIError(callErr)
+		}
+		return list, nil
+	})
+	if err != nil {
+		return false, fmt.Errorf("list issue comments for %s#%d: %w", repo, pr, err)
+	}
+	for _, comment := range comments {
+		if strings.Contains(comment.GetBody(), marker) {
+			return true, nil
+		}
+	}
+	return false, nil
+}
+
 // GetIssueComment returns the current body of an existing PR/issue comment.
 func (ic *InstallationClient) GetIssueComment(ctx context.Context, repo string, commentID int64) (string, error) {
 	owner, repoName := splitRepo(repo)

pkg/metrics/metrics.go

@@ -1520,6 +1520,10 @@ const (
 	// MergeGateSourceSweep marks a request recorded by the backstop sweep
 	// over recently completed applies.
 	MergeGateSourceSweep = "sweep"
+	// MergeGateSourceReleaseSweep marks a settle request backfilled by the
+	// sweep over terminal applies whose preflight held sibling checks but
+	// whose settle was never recorded.
+	MergeGateSourceReleaseSweep = "release_sweep"
 )
 
 // RecordMergeGateRecorded counts durable merge gate requests recorded
@@ -1602,3 +1606,14 @@ func RecordMergeGateTerminatedStuck(ctx context.Context, terminated int64) {
 		"Total merge gate requests terminated by the stuck-processing sweep", "{request}",
 	)
 }
+
+// RecordMergeGatePreflightRearmed counts terminally failed preflight renders
+// re-armed because their apply is still active. A sustained rate means the
+// code host keeps rejecting the hold rendering (outage, auth failure) while
+// applies run on stored holds — sibling PRs' visible checks stay stale until
+// a render lands, so find the failing render in the merge gate logs.
+func RecordMergeGatePreflightRearmed(ctx context.Context, reopened int64) {
+	addCounterN(ctx, reopened, "schemabot.merge_gate.preflight_renders_rearmed_total",
+		"Total terminally failed preflight renders re-armed for still-active applies", "{request}",
+	)
+}

pkg/webhook/check_runs.go

@@ -119,6 +119,17 @@ var schemaChangedReplanFailedBlock = checkBlockReason{
 	message:        "The live schema for this database changed after this plan was computed, and SchemaBot could not re-plan the PR against it. Re-run `schemabot plan` (or push a new commit) before this check can pass; see server logs for the re-plan failure.",
 }
 
+// applyInFlightBlock is used while an apply is running against a check's
+// target: the stored verdict was computed against the pre-apply live schema,
+// so a merge must not land on it before the apply finishes. The preflight
+// fan-out writes it before the apply's engine work starts, and the apply's
+// settle fan-out re-plans the check against the resulting schema, which
+// replaces this hold with a live verdict.
+var applyInFlightBlock = checkBlockReason{
+	blockingReason: "apply_in_flight_on_target",
+	message:        "An apply is currently changing this database's live schema, so this check is held until it finishes. SchemaBot then re-plans this PR against the resulting schema and refreshes this check automatically.",
+}
+
 // noAllowedConfiguredEnvironmentsBlock is used when schema files changed but
 // the server-configured environments for the database do not overlap this
 // service's allowed_environments. SchemaBot cannot safely plan the schema

pkg/webhook/handler.go

@@ -363,11 +363,6 @@ func NewHandlerWithDispatch(service *api.Service, ghClients github.ClientSet, we
 			return nil
 		}
 
-		// Wake the merge gate processor as soon as a drive tail records a
-		// request, so sibling PR checks re-plan without waiting for the next
-		// poll tick. The durable request row stays the source of truth: a
-		// kick lost to a pod boundary only costs poll latency.
-		service.OnMergeGateRecorded = h.KickMergeGate
 	}
 
 	return h