diff --git a/scripts/telemetry_adoption_report.py b/scripts/telemetry_adoption_report.py
index 0e675b53e8..de8e722a96 100644
--- a/scripts/telemetry_adoption_report.py
+++ b/scripts/telemetry_adoption_report.py
@@ -2279,13 +2279,18 @@ def pulse_intelligence_row_analysis_keys(
 ) -> tuple[set[str], set[str]]:
     cohort_keys: set[str] = set()
     signal_groups: set[str] = set()
-    for field in PULSE_INTELLIGENCE_ANALYSIS_BOOL_FIELDS:
-        if not parse_optional_bool(row.get(field)):
+    # Older schemas and sparse exports omit most signals. Intersect in C rather
+    # than parsing every absent field for every heartbeat; explicit nulls have
+    # the same no-signal meaning. Keep the existing parsers for supplied values.
+    for field in row.keys() & PULSE_INTELLIGENCE_COHORT_BOOL_KEYS_BY_FIELD.keys():
+        value = row[field]
+        if value is None or not parse_optional_bool(value):
             continue
         cohort_keys.update(PULSE_INTELLIGENCE_COHORT_BOOL_KEYS_BY_FIELD[field])
         signal_groups.update(PULSE_INTELLIGENCE_SIGNAL_GROUP_BOOL_KEYS_BY_FIELD[field])
-    for field in PULSE_INTELLIGENCE_ANALYSIS_COUNT_FIELDS:
-        if parse_optional_nonnegative_int(row.get(field)) <= 0:
+    for field in row.keys() & PULSE_INTELLIGENCE_COHORT_COUNT_KEYS_BY_FIELD.keys():
+        value = row[field]
+        if value is None or parse_optional_nonnegative_int(value) <= 0:
             continue
         cohort_keys.update(PULSE_INTELLIGENCE_COHORT_COUNT_KEYS_BY_FIELD[field])
         signal_groups.update(PULSE_INTELLIGENCE_SIGNAL_GROUP_COUNT_KEYS_BY_FIELD[field])
diff --git a/scripts/tests/test_telemetry_adoption_report.py b/scripts/tests/test_telemetry_adoption_report.py
index 66511111fa..f7e91fe0b6 100644
--- a/scripts/tests/test_telemetry_adoption_report.py
+++ b/scripts/tests/test_telemetry_adoption_report.py
@@ -2120,6 +2120,76 @@ class TelemetryAdoptionReportTest(unittest.TestCase):
                     expected_signal_conversion,
                 )
 
+    def test_pulse_intelligence_sparse_rows_parse_only_supplied_signals(self) -> None:
+        row = {
+            "install_id": "synthetic",
+            "received_at": "2026-07-23 12:00:00",
+            "paid_license": False,
+            "pulse_intelligence_loop_configured": "yes",
+            "pulse_intelligence_patrol_new_findings_30d": "2",
+            "pulse_intelligence_approved_action_successes_30d": "invalid",
+            "unrelated": 99,
+        }
+        # Full projections include explicit nulls; sparse older exports omit
+        # them. Neither should spend parser calls on unavailable signals.
+        projected = dict.fromkeys(report.REPORT_ROW_COLUMNS)
+        projected.update(row)
+        for supplied in (row, projected):
+            with self.subTest(projected=len(supplied) > len(row)):
+                with mock.patch.object(
+                    report, "parse_optional_bool", wraps=report.parse_optional_bool
+                ) as parse_bool, mock.patch.object(
+                    report,
+                    "parse_optional_nonnegative_int",
+                    wraps=report.parse_optional_nonnegative_int,
+                ) as parse_count:
+                    actual = report.pulse_intelligence_row_analysis_keys(supplied)
+                self.assertEqual(parse_bool.call_args_list, [mock.call("yes")])
+                self.assertCountEqual(
+                    parse_count.call_args_list, [mock.call("2"), mock.call("invalid")]
+                )
+                expected_cohorts = {
+                    key for key, _, bool_fields, count_fields
+                    in report.PULSE_INTELLIGENCE_OUTCOME_COHORTS
+                    if report.pulse_intelligence_row_matches_cohort(
+                        row, bool_fields, count_fields
+                    )
+                }
+                expected_groups = report.pulse_intelligence_row_signal_groups(row)
+                self.assertEqual(actual[0], expected_cohorts)
+                self.assertEqual(
+                    report.pulse_intelligence_derive_signal_groups(actual[1]),
+                    expected_groups,
+                )
+
+    def test_pulse_intelligence_sparse_analysis_preserves_all_field_coercions(self) -> None:
+        values = (None, True, False, 0, 1, -1, 0.5, "2", "yes", "no", " ", "bad", [], {})
+        fields = set(report.PULSE_INTELLIGENCE_ANALYSIS_BOOL_FIELDS) | set(
+            report.PULSE_INTELLIGENCE_ANALYSIS_COUNT_FIELDS
+        )
+        # The independent helpers walk the declared cohorts/groups, not the
+        # optimized field maps. Cover every field, including group-only fields.
+        for field in sorted(fields):
+            for value in values:
+                with self.subTest(field=field, value=value):
+                    row = {field: value, "undeclared_signal": True}
+                    expected_cohorts = {
+                        key for key, _, bool_fields, count_fields
+                        in report.PULSE_INTELLIGENCE_OUTCOME_COHORTS
+                        if report.pulse_intelligence_row_matches_cohort(
+                            row, bool_fields, count_fields
+                        )
+                    }
+                    expected_groups = report.pulse_intelligence_row_signal_groups(row)
+                    actual_cohorts, actual_groups = (
+                        report.pulse_intelligence_row_analysis_keys(row)
+                    )
+                    self.assertEqual(actual_cohorts, expected_cohorts)
+                    self.assertEqual(
+                        report.pulse_intelligence_derive_signal_groups(actual_groups),
+                        expected_groups,
+                    )
+
     def test_pulse_intelligence_production_scale_performance_guard(self) -> None:
         now = datetime(2026, 7, 23, 12, tzinfo=timezone.utc)
         rows: list[dict[str, object]] = []