diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 91b9d0238..6f1716886 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -158,10 +158,13 @@ jobs:
             # branch for a pull request, the previous head for a master push,
             # master for a manual dispatch. Any raised limit, widened tolerance
             # or slack, or removed entry fails, unless a maintainer put the
-            # perf-baseline-increase label on the pull request (for a master
-            # push: on the pull request merged as the pushed commit). Labels
-            # are read from the API, not the event payload, so re-running this
-            # job after adding the label picks it up.
+            # perf-baseline-increase label on the pull request. For a master
+            # push the label counts only when the push added exactly one
+            # first-parent commit (a squash or merge of one PR) and that
+            # commit's PR carries it: the check compares the whole push, so a
+            # multi-commit push cannot borrow one PR's label for another's
+            # increase. Labels are read from the API, not the event payload,
+            # so re-running this job after adding the label picks it up.
             - name: Refuse baseline increases against the previous revision
               env:
                   EVENT_NAME: ${{ github.event_name }}
@@ -196,7 +199,13 @@ jobs:
                           labels="$(gh api "repos/$REPOSITORY/issues/$PR_NUMBER/labels" --jq '.[].name')"
                           ;;
                       push)
-                          labels="$(gh api "repos/$REPOSITORY/commits/$HEAD_SHA/pulls" --jq '.[].labels[].name')"
+                          parent="$(gh api "repos/$REPOSITORY/commits/$HEAD_SHA" --jq '.parents[0].sha')"
+                          if [ "$parent" = "$BEFORE_SHA" ]; then
+                              labels="$(gh api "repos/$REPOSITORY/commits/$HEAD_SHA/pulls" --jq '.[].labels[].name')"
+                          else
+                              echo "This push added more than one commit; the $INCREASE_LABEL label is not consulted."
+                              labels=""
+                          fi
                           ;;
                       *)
                           labels=""
diff --git a/docs/architecture/performance-journeys.md b/docs/architecture/performance-journeys.md
index b411d40ba..fbc445d14 100644
--- a/docs/architecture/performance-journeys.md
+++ b/docs/architecture/performance-journeys.md
@@ -358,7 +358,9 @@ The job also refuses a weakened baselines file:
 `master` for a manual dispatch) and fails when any
 entry's enforced limit (`value × toleranceRatio` or `value + slack`) went up,
 a tolerance or slack widened or an entry disappeared, so a PR cannot grow the
-payload and raise the baseline to match. Lowered limits and new entries pass.
+payload and raise the baseline to match. A counter's `value` may not go up
+either, even when narrower slack lowers its limit. Lowered limits and new
+entries pass.
 
 Baselines only move down. Lower `value` in the same PR as the change that
 earned it, set `updatedAt` and `evidencePr`, and paste the measurement output
@@ -369,8 +371,10 @@ per-file breakdown, and ask a maintainer to add the `perf-baseline-increase`
 label. With the label the direction check prints the weakened entries as
 `ALLOWED` and passes; the job reads labels from the API when it runs, so
 re-run the job after the label is added. For a `master` push the label is
-read from the pull request merged as the pushed commit, so a squash merge of
-a labelled PR passes and a direct push that raises a baseline still fails.
+read from the pull request merged as the pushed commit, and only when the
+push added exactly one first-parent commit: a squash or merge of a labelled
+PR passes, while a direct push, or a push of several commits (which the check
+compares as a whole), that raises a baseline still fails.
 Only people with triage access can set labels, so the label is the
 maintainer decision.
 
diff --git a/tools/performance/check-baseline-direction.mjs b/tools/performance/check-baseline-direction.mjs
index e17b150a2..b0b8b9949 100644
--- a/tools/performance/check-baseline-direction.mjs
+++ b/tools/performance/check-baseline-direction.mjs
@@ -7,7 +7,9 @@
  * check closes that gap: given the baselines file of the target branch and
  * the one of the PR, any entry whose enforced limit (`value × toleranceRatio`
  * or `value + slack`) went up, whose tolerance or slack widened, or that
- * disappeared, is a failure. New entries and lowered limits pass.
+ * disappeared, is a failure. A counter's `value` may not go up either, even
+ * when narrower slack lowers its limit: the value is the measured evidence.
+ * New entries and lowered limits pass.
  *
  * `--allow-increase` turns those failures into printed "allowed" lines. CI
  * passes it only when a maintainer put the perf-baseline-increase label on
@@ -97,6 +99,13 @@ export function compareBaselineDirection({
             weakened.push(
                 `${label}: baseline raised from ${formatNumber(baseLimit)} to ${formatNumber(headLimit)}${unit}. Baselines only move down; bring the measurement back under ${formatNumber(baseLimit)}, or make the case for the increase in the PR and ask a maintainer to add the ${BASELINE_INCREASE_LABEL} label.`
             );
+        } else if (
+            headEntry.toleranceRatio === undefined &&
+            headEntry.value > baseEntry.value
+        ) {
+            weakened.push(
+                `${label}: baseline value raised from ${formatNumber(baseEntry.value)} to ${formatNumber(headEntry.value)}${unit} while slack narrowed from ${formatNumber(baseSlack)} to ${formatNumber(headSlack)}. A counter's value only moves down; narrowing its slack does not offset a raise. Ask a maintainer to add the ${BASELINE_INCREASE_LABEL} label if the raise is deliberate.`
+            );
         } else if (headLimit < baseLimit) {
             result.lowered.push(
                 `${label}: ${formatNumber(baseLimit)} -> ${formatNumber(headLimit)}${unit}.`
diff --git a/tools/performance/check-baseline-direction.test.mjs b/tools/performance/check-baseline-direction.test.mjs
index 87b3ba9e6..54b1346a2 100644
--- a/tools/performance/check-baseline-direction.test.mjs
+++ b/tools/performance/check-baseline-direction.test.mjs
@@ -184,6 +184,35 @@ test('a widened or newly added slack fails; a narrowed one lowers the limit', ()
     assert.match(raised.failures[0], /perf-baseline-increase label/);
 });
 
+test('a counter value raised behind narrower slack still fails', () => {
+    const withSlack = (value, slack) => ({
+        version: 1,
+        journeys: {
+            launch: {
+                'renderer.initialBytes': { value, unit: 'bytes', slack },
+            },
+        },
+    });
+    const result = compareBaselineDirection({
+        base: withSlack(100, 10),
+        head: withSlack(105, 0),
+    });
+    assert.equal(result.failures.length, 1);
+    assert.deepEqual(result.lowered, []);
+    assert.match(
+        result.failures[0],
+        /renderer\.initialBytes: baseline value raised from 100 to 105 bytes while slack narrowed from 10 to 0/
+    );
+
+    const allowed = compareBaselineDirection({
+        base: withSlack(100, 10),
+        head: withSlack(105, 0),
+        allowIncrease: true,
+    });
+    assert.deepEqual(allowed.failures, []);
+    assert.equal(allowed.allowed.length, 1);
+});
+
 test('allowIncrease reports every weakening as allowed instead of failing', () => {
     const result = compareBaselineDirection({
         base: file(100, { cdTicks: { value: 5 } }),