feat(calibration): consume the ai_review_close_confidence override and flip the knob to live (#8197)

The #8176 gate-authority change, mirroring the satisfaction floor's
consumption discipline exactly:

- One bounds-validated resolution point: getKnobOverride reads the knob's
  system_flags row only when its own wrangler var
  (AI_REVIEW_CLOSE_CONFIDENCE_AUTOTUNE_ENABLED, false by default) is ON, and
  refuses anything not strictly below the shipped 0.93 or under the 0.85
  hard minimum. gateCheckPolicy threads it as the LAST-resort default — an
  explicit per-repo gate.aiReview.closeConfidence setting always wins.
- Generic apply machinery (knob-loosening-run.ts): the satisfaction
  run/status/tick generalized over the registry entry rather than
  duplicated. The satisfaction floor keeps its legacy module and event
  shape; the shared cron tick now runs every OTHER live knob, double-gated
  per knob. Report-only refusal stays enforced in the write path itself.
- Registry entries carry their apply plumbing (override flag key, event
  type, autotune var), pinned against the legacy constants by invariant
  tests; the close-confidence knob flips to applyMode live.
- Operator surface: GET /v1/internal/calibration/knobs renders every live
  knob's flag state, shipped/live/override values, and applied history
  through one projector that reads both proposal spellings.

Closes #8176

Author: JSONbored

src/api/routes.ts

@@ -321,6 +321,7 @@ import { getPublicStats, isPublicStatsEnabled, resolvePublicStatsManifestOverrid
 import { loadPublicAccuracyTrend } from "../services/public-accuracy-trend";
 import { loadCalibrationTrend } from "../services/rule-calibration-trend";
 import { isSatisfactionFloorAutotuneEnabled, loadSatisfactionFloorStatus, runSatisfactionFloorLoosening } from "../services/satisfaction-floor-loosening-run";
+import { loadLiveKnobStatuses } from "../services/knob-loosening-run";
 import { loadPublicReuseRateTrend } from "../services/public-reuse-rate-trend";
 import { loadPublicReviewVolumeTrend } from "../services/public-review-volume-trend";
 import { buildMaintainerQualityDashboard, isMaintainerQualityDataStale } from "../services/maintainer-quality-dashboard";
@@ -4832,6 +4833,11 @@ export function createApp() {
   // off. Same INTERNAL_JOB_TOKEN gate via the /v1/internal/* middleware; aggregate numbers/verdicts only.
   app.get("/v1/internal/calibration/satisfaction-floor", async (c) => c.json(await loadSatisfactionFloorStatus(c.env)));
 
+  // The #8161 surface generalized across EVERY live registry knob (#8176): one endpoint, one projector,
+  // per-knob flag state + shipped/live/override values + applied history (both split verdicts). Same
+  // deliberate non-flag-gating and INTERNAL_JOB_TOKEN posture as the satisfaction-floor read above.
+  app.get("/v1/internal/calibration/knobs", async (c) => c.json({ knobs: await loadLiveKnobStatuses(c.env) }));
+
   app.post("/v1/internal/jobs/refresh-registry", async (c) => {
     const message: JobMessage = { type: "refresh-registry", requestedBy: "api" };
     await c.env.JOBS.send(message);

src/queue/gate-checks.ts

@@ -66,6 +66,10 @@ export function gateCheckPolicy(
     guardrailHit: boolean;
     guardrailMatches?: ReturnType<typeof guardrailPathMatches> | undefined;
   },
+  // #8176: the backtest-gated GLOBAL default-override for the AI close-confidence floor, resolved by the
+  // env-bearing caller (getAiReviewCloseConfidenceOverride — flag-gated + bounds-validated). It only fills
+  // the DEFAULT: an explicit per-repo `gate.aiReview.closeConfidence` setting always wins below.
+  aiReviewCloseConfidenceOverride?: number | null,
 ) {
   // `settings` is already the EFFECTIVE config (`.loopover.yml` > DB > defaults), resolved upstream by
   // resolveRepositorySettings, so the blocker modes here reflect the repo's config file directly.
@@ -81,8 +85,9 @@ export function gateCheckPolicy(
     qualityGateMinScore: settings.qualityGateMinScore ?? null,
     aiReviewGateMode: settings.aiReviewMode,
     // Calibrated AI close-confidence floor (#7) — config-as-code via `.loopover.yml gate.aiReview.closeConfidence`,
-    // resolved into settings upstream. `null`/undefined ⇒ advisory.ts applies the 0.93 default.
-    aiReviewCloseConfidence: settings.aiReviewCloseConfidence ?? null,
+    // resolved into settings upstream. When the repo has no explicit setting, the #8176 backtest-gated
+    // global override (if any) becomes the default; `null` ⇒ advisory.ts applies the 0.93 shipped default.
+    aiReviewCloseConfidence: settings.aiReviewCloseConfidence ?? aiReviewCloseConfidenceOverride ?? null,
     // Sub-floor AI-judgment disposition (#4603) — DB-backed (dashboard-settable) + `.loopover.yml
     // gate.aiReview.lowConfidenceDisposition` override, resolved into settings upstream. `null`/undefined ⇒
     // advisory.ts applies the "hold_for_review" default.

src/queue/job-dispatch.ts

@@ -31,6 +31,7 @@ import { isPrReconciliationEnabled, resolvePrReconciliationManifestOverride, run
 import { isActiveReviewReconciliationEnabled, resolveActiveReviewReconciliationManifestOverride, runActiveReviewReconciliation } from "../review/active-review-reconciliation";
 import { isSelfTuneEnabled, runSelfTune } from "../review/selftune-wire";
 import { isSatisfactionFloorAutotuneEnabled, runScheduledSatisfactionFloorLoosening } from "../services/satisfaction-floor-loosening-run";
+import { GENERIC_LIVE_KNOBS, isKnobAutotuneEnabled, runScheduledKnobLoosening } from "../services/knob-loosening-run";
 import { runSelfTuneBreaker } from "../review/outcomes-wire";
 import { isRagEnabled } from "../review/rag-wire";
 import { processSubmitDraft } from "../services/draft";
@@ -348,6 +349,11 @@ export async function processJob(env: Env, message: JobMessage): Promise<void> {
       // flag is ON, but a stale in-flight job landing after a flag-flip must still no-op. Never throws into
       // the queue (the scheduled wrapper fails safe).
       if (isSatisfactionFloorAutotuneEnabled(env)) await runScheduledSatisfactionFloorLoosening(env);
+      // #8176: every LATER live registry knob rides the same tick through the generic runner — each knob
+      // is double-gated on its OWN wrangler var, so an un-flagged knob does zero work here.
+      for (const knob of GENERIC_LIVE_KNOBS) {
+        if (isKnobAutotuneEnabled(env, knob)) await runScheduledKnobLoosening(env, knob);
+      }
       return;
     case "selftune":
       // Convergence (self-improve / auto-tune, flag LOOPOVER_REVIEW_SELFTUNE). Defense-in-depth: the cron only

src/queue/processors.ts

@@ -426,6 +426,7 @@ export { runRetentionPrune } from "./retention";
 // test/unit/gate-check-policy.test.ts and test/unit/repository-settings-enforcement.test.ts's existing
 // `import { gateCheckPolicy } from "../../src/queue/processors"` keeps working unchanged.
 import { auditGateCheckPermissionMissing, gateCheckPolicy, recordPublishedGateCheckSummary } from "./gate-checks";
+import { getAiReviewCloseConfidenceOverride } from "../services/knob-loosening-run";
 export { gateCheckPolicy } from "./gate-checks";
 // #4013 step 9: same shim shape for the AI-review-orchestration functions -- imported here for this file's
 // own remaining internal callers, and re-exported so the many existing tests importing claimAiReviewLock,
@@ -1538,6 +1539,9 @@ export async function sweepRepoRegate(
   // isScheduledRegateSweepJob (queue-common.ts) misclassifies it as background maintenance and it inherits the
   // exact starvation this priority mechanism exists to avoid. Ordinary stale candidates keep the sweep prefix
   // unchanged.
+  // #8176: the global close-confidence default-override, resolved once for the sweep (same value the main
+  // webhook path threads; an explicit per-repo setting still wins inside gateCheckPolicy).
+  const sweepCloseConfidenceOverride = await getAiReviewCloseConfidenceOverride(env);
   for (const [index, pr] of candidates.entries()) {
     const others = openPullRequests.filter(
       (other) => other.number !== pr.number,
@@ -1561,7 +1565,7 @@ export async function sweepRepoRegate(
     });
     const gate = evaluateGateCheck(
       advisory,
-      gateCheckPolicy(settings, null, undefined, pr.slopRisk ?? null),
+      gateCheckPolicy(settings, null, undefined, pr.slopRisk ?? null, undefined, undefined, sweepCloseConfidenceOverride),
     );
     verdicts[String(pr.number)] = gate.conclusion;
     if (gate.conclusion === "failure" || gate.conclusion === "action_required")
@@ -10381,6 +10385,8 @@ async function maybePublishPrPublicSurface(
       slopRisk,
       authorHistory,
       gateSizeContext,
+      // #8176: backtest-gated global default for the close-confidence floor (explicit per-repo wins inside).
+      await getAiReviewCloseConfidenceOverride(env),
     );
     gateEvaluation = await withReviewPipelineSpan(
       "selfhost.review.gate",
@@ -11871,7 +11877,7 @@ async function maybeProcessResolveCommand(env: Env, deliveryId: string, payload:
   if (!findingRef.ok) { await recordAuditEvent(env, { eventType: "github_app.finding_resolved_skipped", actor: req.actor, targetKey, outcome: "completed", detail: findingRef.reason, metadata: { deliveryId, repoFullName: req.repoFullName, reason: findingRef.reason } }); await recordGithubProductUsage(env, "finding_resolved_skipped", { actor: req.actor, repoFullName: req.repoFullName, targetKey, outcome: "skipped", metadata: { reason: findingRef.reason } }); return true; }
   const { advisory } = await buildAuthorizedPrActionAdvisory(env, req.repoFullName, pr, settings);
   await appendPublishedAiReviewFindingsForResolve(env, req.repoFullName, pr, settings.aiReviewMode, advisory);
-  const gate = evaluateGateCheck(advisory, gateCheckPolicy(settings, null, undefined, pr.slopRisk ?? null));
+  const gate = evaluateGateCheck(advisory, gateCheckPolicy(settings, null, undefined, pr.slopRisk ?? null, undefined, undefined, await getAiReviewCloseConfidenceOverride(env)));
   const selection = selectWarningsForResolve(gate.warnings, findingRef);
   if (selection.reason === "finding_not_found") { await recordAuditEvent(env, { eventType: "github_app.finding_resolved_skipped", actor: req.actor, targetKey, outcome: "completed", detail: selection.reason, metadata: { deliveryId, repoFullName: req.repoFullName, reason: selection.reason } }); await recordGithubProductUsage(env, "finding_resolved_skipped", { actor: req.actor, repoFullName: req.repoFullName, targetKey, outcome: "skipped", metadata: { reason: selection.reason } }); return true; }
   const mode = resolveAgentActionMode({ globalPaused: isGlobalAgentPause(env) || (await isGlobalAgentFrozen(env)), agentPaused: settings.agentPaused, agentDryRun: settings.agentDryRun });
@@ -12102,7 +12108,7 @@ async function maybeProcessExplainCommand(env: Env, deliveryId: string, payload:
   }
   const { advisory } = await buildAuthorizedPrActionAdvisory(env, req.repoFullName, pr, settings);
   await appendPublishedAiReviewFindingsForResolve(env, req.repoFullName, pr, settings.aiReviewMode, advisory);
-  const gate = evaluateGateCheck(advisory, gateCheckPolicy(settings, null, undefined, pr.slopRisk ?? null));
+  const gate = evaluateGateCheck(advisory, gateCheckPolicy(settings, null, undefined, pr.slopRisk ?? null, undefined, undefined, await getAiReviewCloseConfidenceOverride(env)));
   const selection = selectWarningsForResolve(gate.warnings, findingRef);
   if (selection.reason === "finding_not_found") {
     const notFound = sanitizePublicComment([AGENT_COMMAND_COMMENT_MARKER, "", "> [!NOTE]", `> **No review finding \`${findingRef.findingCode}\` on this PR**`, "> That id is not among this PR's current review findings — re-run `@loopover explain <finding-id>` with an id from the review summary.", "", "---", loopoverFooter(env)].join("\n"));