:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.
Python ab-testingai-analyticsanalyticscdpdata-warehouseexperimentsfeature-flagsjavascriptproduct-analyticspythonreactsession-replaysurveystypescriptweb-analytics
fix(signals): exclude previous-period lines from anomaly scoring (#123037)
posthog[bot]73b8eb348a78
diff --git a/products/signals/backend/anomaly_scores/scoring.py b/products/signals/backend/anomaly_scores/scoring.pyindex 25f9badba1d1..990bac80ace1 100644--- a/products/signals/backend/anomaly_scores/scoring.py+++ b/products/signals/backend/anomaly_scores/scoring.py@@ -8,6 +8,7 @@ from typing import Any, castfrom zoneinfo import ZoneInfoimport numpy as np+import structlogfrom posthog.schema import BreakdownFilter, ChartDisplayType, DateRange, IntervalType, TrendsQuery@@ -38,6 +39,8 @@ from products.signals.backend.anomaly_scores.config import (resolve_profile,)+logger = structlog.get_logger(__name__)+# Naming fields do not change what a series counts, so a rename keeps its key._SERIES_NAMING_FIELDS = {"custom_name", "name"}@@ -197,6 +200,9 @@ def build_scoring_query(query: TrendsQuery, buckets: int, limits: ScoringLimits"dateRange": DateRange(date_from=f"-{buckets}{_DATE_UNITS[_interval(query)]}", daysOfWeek=days_of_week),"compareFilter": None,}+ if query.trendsFilter and query.trendsFilter.display == ChartDisplayType.METRIC:+ # The runner adds the previous period for a Metric display. A line display returns the same buckets.+ update["trendsFilter"] = query.trendsFilter.model_copy(update={"display": ChartDisplayType.ACTIONS_LINE_GRAPH})if has_breakdown_filter(query.breakdownFilter):lines_per_value = len(_formulas(query)) or len(query.series)breakdown_filter = query.breakdownFilter or BreakdownFilter()@@ -316,9 +322,20 @@ def score_insight(if not isinstance(results, list):raise RuntimeError(f"No results for insight {insight.pk}")+ current_results = [+ result for result in cast(list[dict[str, Any]], results) if result.get("compare_label") != "previous"+ ]+ if len(current_results) < len(results):+ # A previous-period line has the key of its current line and old buckets, so it would write a false series.+ logger.warning(+ "anomaly_scores.previous_period_skipped",+ insight_id=insight.pk,+ skipped=len(results) - len(current_results),+ )+series: list[ScoredSeries] = []key_counts: dict[str, int] = {}- for result in cast(list[dict[str, Any]], results)[: limits.max_series]:+ for result in current_results[: limits.max_series]:buckets = [_parse_bucket(day, timezone) for day in result.get("days") or []]breakdown_value = _breakdown_label(result.get("breakdown_value"))key = series_key(insight.pk, _series_definition(scoring_query, result), breakdown_value)diff --git a/products/signals/backend/test/test_anomaly_scores.py b/products/signals/backend/test/test_anomaly_scores.pyindex 8a12923a13a6..9b9dcc60337d 100644--- a/products/signals/backend/test/test_anomaly_scores.py+++ b/products/signals/backend/test/test_anomaly_scores.py@@ -11,6 +11,7 @@ from parameterized import parameterizedfrom posthog.schema import (ChartDisplayType,+ CompareFilter,EventPropertyFilter,EventsNode,FunnelsQuery,@@ -221,6 +222,19 @@ class TestScoreInsight(SimpleTestCase):assert named["web"] == single["web"]assert named["app"] != named["web"]+ def test_previous_period_lines_are_not_scored_and_the_current_line_keeps_its_key(self) -> None:+ pageview = EventsNode(event="$pageview")+ line = self._keys(_insight([pageview]), [_result(0, "pageview")])+ metric = self._keys(+ _insight([pageview], trendsFilter=TrendsFilter(display=ChartDisplayType.METRIC)),+ [+ {**_result(0, "pageview"), "compare_label": "current"},+ {**_result(0, "pageview"), "compare_label": "previous"},+ ],+ )++ assert metric == line+@parameterized.expand([("one_query", 1), ("one_full_wave", 10), ("two_waves", 11), ("many_waves", 120)])def test_every_query_wave_of_an_insight_stops_before_the_activity_timeout(self, _name: str, queries: int) -> None:limits: list[int] = []@@ -242,6 +256,27 @@ class TestScoreInsight(SimpleTestCase):assert limits[0] == 150+class TestMetricScoringQuery(SimpleTestCase):+ @parameterized.expand([("without_compare", None), ("with_saved_compare", CompareFilter(compare=True))])+ def test_metric_display_is_scored_as_a_line_without_the_previous_period(+ self, _name: str, compare_filter: CompareFilter | None+ ) -> None:+ insight = _insight(+ [EventsNode(event="$pageview"), EventsNode(event="signed_up")],+ trendsFilter=TrendsFilter(display=ChartDisplayType.METRIC),+ compareFilter=compare_filter,+ )++ with patch("products.signals.backend.anomaly_scores.scoring.TrendsQueryRunner") as runner:+ runner.return_value.run.return_value = MagicMock(results=[])+ score_insight(insight, Team(timezone="UTC"), now=datetime(2026, 1, 10, tzinfo=UTC))++ query = runner.call_args.kwargs["query"]+ assert query.trendsFilter.display == ChartDisplayType.ACTIONS_LINE_GRAPH+ assert query.compareFilter is None+ assert query_cost(insight, MagicMock(spec=Team)) == 2++class TestFetchBuckets(SimpleTestCase):def test_unbounded_smoothing_is_not_scored(self) -> None:query = TrendsQuery(