Skip to content

: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.

fix(alerts): check same-weekday baseline before true positive (#122235)

posthog[bot]e92b9a727bdc
diff --git a/posthog/temporal/ai/anomaly_investigation/prompts.py b/posthog/temporal/ai/anomaly_investigation/prompts.py
index a665625307fd..b341d218c0e6 100644
--- a/posthog/temporal/ai/anomaly_investigation/prompts.py
+++ b/posthog/temporal/ai/anomaly_investigation/prompts.py
@@ -101,6 +101,15 @@ Magnitude check (do this before classifying):
seasonality, burstiness, occasional outlier buckets. Sense-check the
firing against the series' broader shape, not just the triggered point
and its immediate neighbours.
+- Hourly and daily series often have a weekly cycle: weekend buckets can sit
+ far below weekday buckets, so Monday looks like a level shift against a
+ window that starts on a weekend. Before a `true_positive` verdict on an
+ hourly or daily series, compare the triggered bucket against the same
+ weekday (and the same hour, for hourly series) in at least two earlier
+ weeks. Call `fetch_metric_series` with a `date_from` that covers those
+ weeks. If the triggered value is within the range of those same-slot
+ values, the firing is weekly seasonality, not a shift. Name the
+ same-slot values you compared in the evidence.
- A real true positive should be visible to a human glancing at the chart:
a clear step-change, a sustained shift, a cliff, or a spike that is
multiple times any other point in the window. If you have to squint, it
@@ -113,8 +122,9 @@ Verdict rubric:
tied to a known release or property change.
- `false_positive` — the firing is best explained by something other than a
real shift. Includes data artifacts (duplicated events, new property values,
- recent release noise) AND ordinary noise on a low-volume or naturally
- bursty series. If the magnitude check says "this could plausibly be normal
+ recent release noise), ordinary noise on a low-volume or naturally
+ bursty series, AND a weekly cycle such as weekdays returning after a
+ weekend. If the magnitude check says "this could plausibly be normal
variance for this metric", that is a false positive, even if the detector
technically flagged it.
- `inconclusive` — not enough evidence to call it either way within the