ci: make performance gate runner-noise aware

.github/workflows/validate.yml

@@ -1,6 +1,10 @@
 name: Validate
 on:
   push:
+    # The branch commit is the release-gate authority. An annotated tag points
+    # to the same SHA and must not duplicate every expensive browser/perf job.
+    tags-ignore:
+      - '**'
   pull_request:
   schedule:
     - cron: "0 4 * * 1"

demo/performance/README.md

@@ -34,7 +34,8 @@ not normal-performance targets; the base-relative comparison catches smaller
 regressions. Small fast operations receive an absolute noise
 allowance so normal scheduler jitter does not become a false regression. Heap,
 Long Tasks, warmed-cache growth and the expected rendered-device count are
-gated separately. Both raw reports and the comparison are always uploaded as
+gated separately. Long-Task maximum/count/total checks use the same
+relative-plus-absolute policy as timings. Both raw reports and the comparison are always uploaded as
 the `large-house-performance` artifact, and the table is written to the GitHub
 job summary.
 
@@ -71,3 +72,10 @@ changing the meaning of `large-house-v1`.
 The `cleanFloor` entry ceiling is 160: the reviewed fixture currently warms
 120 deterministic room/physical-body entries, and the extra 40 slots allow a
 legitimate fixture extension without weakening the separate zero-growth gate.
+
+The absolute switch-cycle/Long-Task ceilings include roughly 20–30% headroom
+over the paired 2026-08-09 Ubuntu run where the unchanged base and candidate
+both reached about 5.3 s / 2.45 s / 22 tasks / 9.9 s total under runner load.
+The same-runner relative checks remain tighter for an actual candidate-only
+regression; this prevents an overloaded but symmetric runner from turning an
+absolute safety ceiling into a flaky code-regression signal.

demo/performance/budgets.json

@@ -49,13 +49,17 @@
       "stat": "median",
       "maxRegressionRatio": 0.35,
       "noiseAllowanceMs": 250,
-      "hardMaxMs": 5000
+      "hardMaxMs": 7000
     }
   },
   "longTasks": {
-    "maxSingleMs": 2000,
-    "maxCountP95": 20,
-    "maxTotalP95Ms": 8000,
+    "maxSingleMs": 3000,
+    "maxSingleRegressionRatio": 0.3,
+    "maxSingleNoiseAllowanceMs": 250,
+    "maxCountP95": 30,
+    "maxCountRegressionRatio": 0.35,
+    "countNoiseAllowance": 3,
+    "maxTotalP95Ms": 12000,
     "maxTotalRegressionRatio": 0.3,
     "noiseAllowanceMs": 150
   },

demo/performance/evaluate.mjs

@@ -95,8 +95,28 @@ export const evaluatePerformanceBudget = ({ candidate, baseline, budgets }) => {
     return windows.length > 0 && windows.every((item) => item?.supported === true);
   });
   checks.push({ id: 'longTask.available', actual: longTasksAvailable ? 1 : 0, limit: 1, pass: longTasksAvailable });
-  checks.push(makeCheck('longTask.maxSingleMs', candidateLong.maxSingleMs, budgets.longTasks.maxSingleMs));
-  checks.push(makeCheck('longTask.countP95', candidateLong.countP95, budgets.longTasks.maxCountP95));
+  const singleRegressionLimit = relativeLimit(
+    baselineLong.maxSingleMs,
+    budgets.longTasks.maxSingleRegressionRatio,
+    budgets.longTasks.maxSingleNoiseAllowanceMs,
+  );
+  checks.push(makeCheck(
+    'longTask.maxSingleMs',
+    candidateLong.maxSingleMs,
+    Math.min(budgets.longTasks.maxSingleMs, singleRegressionLimit),
+    { baseline: baselineLong.maxSingleMs, hardLimit: budgets.longTasks.maxSingleMs },
+  ));
+  const countRegressionLimit = relativeLimit(
+    baselineLong.countP95,
+    budgets.longTasks.maxCountRegressionRatio,
+    budgets.longTasks.countNoiseAllowance,
+  );
+  checks.push(makeCheck(
+    'longTask.countP95',
+    candidateLong.countP95,
+    Math.min(budgets.longTasks.maxCountP95, countRegressionLimit),
+    { baseline: baselineLong.countP95, hardLimit: budgets.longTasks.maxCountP95 },
+  ));
   const longRegressionLimit = relativeLimit(
     baselineLong.totalP95Ms,
     budgets.longTasks.maxTotalRegressionRatio,

test/performance-budget.test.mjs

@@ -13,6 +13,8 @@ const budgets = {
   },
   longTasks: {
     maxSingleMs: 200, maxCountP95: 10, maxTotalP95Ms: 500,
+    maxSingleRegressionRatio: 0.5, maxSingleNoiseAllowanceMs: 20,
+    maxCountRegressionRatio: 0.5, countNoiseAllowance: 2,
     maxTotalRegressionRatio: 0.5, noiseAllowanceMs: 20,
   },
   heap: {
@@ -75,6 +77,17 @@ test('performance budget rejects long tasks, heap growth, cache growth and missi
   );
 });
 
+test('performance budget compares single/count Long Tasks to the same-runner baseline', () => {
+  const baseline = report({ long: 100 });
+  baseline.longTasks.countP95 = 4;
+  const candidate = report({ long: 130 });
+  candidate.longTasks.countP95 = 6;
+  const result = evaluatePerformanceBudget({ baseline, candidate, budgets });
+  assert.equal(result.pass, true);
+  assert.equal(result.checks.find((check) => check.id === 'longTask.maxSingleMs')?.baseline, 100);
+  assert.equal(result.checks.find((check) => check.id === 'longTask.countP95')?.baseline, 4);
+});
+
 test('performance budget refuses incomparable runtime profiles', () => {
   const candidate = report();
   candidate.runtime.chromium = 'different';