fix(web): fleet roll-up success rate is terminal-only, matching the cards [Phase 3c follow-up]

Addresses a Codex finding: the per-tenant cards show the backend success rate,
which is terminal-only (succeeded / succeeded+failed). The fleet roll-up instead
divided by total sessions (which includes active/pending), so the tile and the
cards disagreed — e.g. a tenant with 90 succeeded / 10 failed / 900 active shows
90% on its card but the roll-up showed 9%.

- computeFleetRollup: success rate now sum(succeeded) / (sum(succeeded)+sum(failed)),
  excluding active/pending from the denominator — same definition as DashboardStats.
- successTone: red emphasis now gates on terminal-session count (not total), so a
  tenant with only active sessions reads 0% but isn't flagged as failing.
- Tests: replaced the total-based weighted test with terminal-only weighting and
  added an explicit 90/10/900 case locking in 90% (not 9%).

Web tsc clean; 481 web tests green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/Web/autopilot-monitor-web/app/fleet/lib/__tests__/fleetRollup.test.ts

@@ -37,16 +37,28 @@ describe("computeFleetRollup", () => {
     expect(r.failedLastNDays).toBe(2);
   });
 
-  it("weights the success rate by session volume, not by tenant", () => {
-    // A: 900/1000 = 90%, B: 0/10 = 0%. A naive average would be 45%; the weighted rate is 900/1010 ≈ 89.1%.
+  it("weights the success rate by terminal-session volume, not by tenant", () => {
+    // A: 900 succeeded / 100 failed = 90%, B: 0 succeeded / 10 failed = 0%. A naive average would be 45%;
+    // the weighted terminal rate is 900 / (900 + 100 + 10) = 900/1010 ≈ 89.1%.
     const r = computeFleetRollup([
-      summary({ totalLastNDays: 1000, succeededLastNDays: 900 }),
-      summary({ totalLastNDays: 10, succeededLastNDays: 0 }),
+      summary({ succeededLastNDays: 900, failedLastNDays: 100, totalLastNDays: 1000 }),
+      summary({ succeededLastNDays: 0, failedLastNDays: 10, totalLastNDays: 10 }),
     ]);
     expect(r.successRatePct).toBeCloseTo(89.1, 1);
   });
 
-  it("reports 0% success when the fleet has no sessions", () => {
+  it("excludes active/pending sessions from the success-rate denominator (terminal-only)", () => {
+    // Matches the per-tenant card / backend semantic: 90 succeeded, 10 failed, 900 active. The card shows
+    // 90/(90+10) = 90%; a total-based roll-up would wrongly show 90/1000 = 9%. Lock in terminal-only.
+    const r = computeFleetRollup([
+      summary({ succeededLastNDays: 90, failedLastNDays: 10, activeCount: 900, totalLastNDays: 1000 }),
+    ]);
+    expect(r.successRatePct).toBe(90);
+    expect(r.activeCount).toBe(900);
+    expect(r.totalLastNDays).toBe(1000);
+  });
+
+  it("reports 0% success when the fleet has no terminal sessions", () => {
     const r = computeFleetRollup([summary({ activeCount: 1 }), summary({})]);
     expect(r.totalLastNDays).toBe(0);
     expect(r.successRatePct).toBe(0);

src/Web/autopilot-monitor-web/app/fleet/lib/fleetRollup.ts

@@ -18,14 +18,20 @@ export interface FleetRollup {
   totalLastNDays: number;
   succeededLastNDays: number;
   failedLastNDays: number;
-  /** Weighted success rate across the fleet (sum succeeded / sum total), 1 decimal. 0 when no sessions. */
+  /**
+   * Weighted fleet success rate, 1 decimal. TERMINAL-ONLY: sum(succeeded) / (sum(succeeded) + sum(failed)) —
+   * matches the per-tenant card / backend DashboardStats semantic (succeeded over terminal sessions), so the
+   * roll-up and the cards never disagree. Active/pending sessions are NOT in the denominator. 0 when the
+   * fleet has no terminal sessions.
+   */
   successRatePct: number;
 }
 
 /**
- * Aggregates per-tenant summaries into a fleet roll-up. The success rate is WEIGHTED by session volume
- * (sum of succeeded / sum of total) — not an average of per-tenant rates — so a tenant with 1000 sessions
- * counts more than one with 3. Pure + testable.
+ * Aggregates per-tenant summaries into a fleet roll-up. The success rate is WEIGHTED by terminal-session
+ * volume (sum succeeded / sum terminal) — not an average of per-tenant rates — so a tenant with 1000
+ * terminal sessions counts more than one with 3. Terminal-only matches the backend's per-tenant rate, so a
+ * tenant card and the roll-up tile use the same definition. Pure + testable.
  */
 export function computeFleetRollup(summaries: FleetSummary[]): FleetRollup {
   const acc = summaries.reduce(
@@ -39,10 +45,9 @@ export function computeFleetRollup(summaries: FleetSummary[]): FleetRollup {
     { activeCount: 0, totalLastNDays: 0, succeededLastNDays: 0, failedLastNDays: 0 }
   );
 
+  const terminal = acc.succeededLastNDays + acc.failedLastNDays;
   const successRatePct =
-    acc.totalLastNDays > 0
-      ? Math.round((acc.succeededLastNDays / acc.totalLastNDays) * 1000) / 10
-      : 0;
+    terminal > 0 ? Math.round((acc.succeededLastNDays / terminal) * 1000) / 10 : 0;
 
   return { tenantCount: summaries.length, ...acc, successRatePct };
 }

src/Web/autopilot-monitor-web/app/fleet/page.tsx

@@ -68,7 +68,7 @@ export default function FleetPage() {
         <RollupTile label="Tenants" value={String(myTenants.length)} />
         <RollupTile label="Active sessions" value={String(rollup.activeCount)} />
         <RollupTile label="Failed" value={String(rollup.failedLastNDays)} tone={rollup.failedLastNDays > 0 ? "danger" : "default"} />
-        <RollupTile label="Success rate" value={`${rollup.successRatePct}%`} tone={successTone(rollup.successRatePct, rollup.totalLastNDays)} />
+        <RollupTile label="Success rate" value={`${rollup.successRatePct}%`} tone={successTone(rollup.successRatePct, rollup.succeededLastNDays + rollup.failedLastNDays)} />
       </div>
 
       {myTenants.length === 0 ? (
@@ -88,9 +88,13 @@ export default function FleetPage() {
 
 type Tone = "default" | "danger";
 
-/** Amber/red emphasis only once we actually have sessions to judge — an empty tenant isn't "failing". */
-function successTone(pct: number, total: number): Tone {
-  if (total === 0) return "default";
+/**
+ * Red emphasis only once there are TERMINAL sessions to judge — the success rate is terminal-only
+ * (succeeded / succeeded+failed), so a tenant with only active/pending sessions reads 0% but isn't
+ * "failing"; don't flag it.
+ */
+function successTone(pct: number, terminalCount: number): Tone {
+  if (terminalCount === 0) return "default";
   return pct < 80 ? "danger" : "default";
 }
 
@@ -123,7 +127,7 @@ function FleetCard({ tenant, summary }: { tenant: TenantInfo; summary?: FleetSum
         <Metric
           label="Success"
           value={summary ? `${summary.successRatePct}%` : "—"}
-          danger={!!summary && successTone(summary.successRatePct, summary.totalLastNDays) === "danger"}
+          danger={!!summary && successTone(summary.successRatePct, summary.succeededLastNDays + summary.failedLastNDays) === "danger"}
         />
       </div>
     </div>