Skip to main content
§ 00 · PRODUCT · ANOMALY DETECTION

Six detectors.
One alert queue.

workers/cron/src/insights-detectors.ts runs delivery_quality, coverage_gaps, duplicate_events, identity_gaps, quota_trajectory, and latency_outliers. Schema drift and alert routing turn those findings into operational work.

6
INSIGHT DETECTORS
24h
BASELINE WINDOW
Slack
ALERT ROUTE
cron · insights-detectors.ts · run #1284
delivery_quality: Meta delivered 87.3% vs 96.1% baseline
coverage_gaps: search event absent for 7d
duplicate_events: 0 duplicate event_id collisions in 24h
identity_gaps: fbc_presence_rate down 31pts after deploy
TRUSTED BYNorthfieldGlasshousePeakformHalcyon LabsSable
§ 01 · IMPLEMENTATION REFERENCES

Built from the product surface, not brochure claims.

Detectors
6
delivery, coverage, duplicate, identity, quota, latency
Drift
schema
taxonomy version checks
Alerts
Slack
severity-routed
Evidence
jsonb
queryable finding payloads
// 02 · OPERATING MODEL

What the team actually gets.

ANCHOR SURFACE

Delivery quality

Flags platform delivery rates below target with attempted, delivered, and failed counts.

cron · insights-detectors.ts · run #1284
delivery_quality: Meta delivered 87.3% vs 96.1% baseline
coverage_gaps: search event absent for 7d
duplicate_events: 0 duplicate event_id collisions in 24h
identity_gaps: fbc_presence_rate down 31pts after deploy
LIVE SIGNAL
EMQ 8.7
§ 02

Coverage gaps

Compares actual event names against recommended canonical events for connected platforms.

§ 03

Quota trajectory

Projects month-end event volume before a merchant hits the plan ceiling.

§ 04

Latency outliers

Catches p95 delivery latency jumps by platform and window.

§ 03 · WORKFLOW

From source event to trusted action.

§ 01

Compute

Cron jobs read deliveries, events, billing usage, platform connections, and merchant quota rows.

§ 02

Classify

Findings are severity-graded with evidence, root_cause_hint, and action_type/action_payload.

§ 03

Route

Critical and high issues land in the dashboard and Slack with enough context to fix or snooze.

§ 04 · CAPABILITY CHECK

How it compares to ordinary tracking work.

CAPABILITYTRACKLAYERMANUAL / GTMAPP / DASHBOARD
delivery_qualityattempted/delivered ratemanual dashboardplatform UI
coverage_gapstaxonomy-awaremanual QAlimited
duplicate_eventsevent_id windowtag previewreporting symptom
identity_gapsmatch-key baselinespot checkEMQ view
quota_trajectoryprojected EOMinvoice surpriseplan page
TRACKLAYER
delivery_quality
attempted/delivered rate
coverage_gaps
taxonomy-aware
duplicate_events
event_id window
identity_gaps
match-key baseline
quota_trajectory
projected EOM
MANUAL / GTM
delivery_quality
manual dashboard
coverage_gaps
manual QA
duplicate_events
tag preview
identity_gaps
spot check
quota_trajectory
invoice surprise
APP / DASHBOARD
delivery_quality
platform UI
coverage_gaps
limited
duplicate_events
reporting symptom
identity_gaps
EMQ view
quota_trajectory
plan page
// 05 · FIELD PROOF

Real merchant-shaped cases and measurable signals.

Northfield
87.3%
delivery drop caught before Ads Manager review
Glasshouse
31pts
fbc regression found after checkout migration
Peakform
3d
quota warning before overage
06 · CONSOLE

The references an operator can inspect.

anomaly-detection · implementation hooks
detectDeliveryQuality(merchantId, supabase)
detectCoverageGaps(merchantId, supabase)
detectDuplicateEvents(merchantId, supabase)
detectIdentityGaps(merchantId, supabase)
detectQuotaTrajectory(merchantId, supabase)
detectLatencyOutliers(merchantId, supabase)
§ 07 · MERCHANT SCENARIOS

Where this matters in production.

Peakform · paid launch

A flash sale doubled event volume by day 19.

quota_trajectory projected the overage early enough to upgrade before delivery was constrained.

Glasshouse · migration

Checkout moved domains and stopped preserving fbc.

identity_gaps alerted with fbc_presence_rate evidence and affected events.

Sable · international store

TikTok p95 delivery latency climbed after a regional DNS change.

latency_outliers isolated the platform and time window for engineering.

// 08 · FAQ

Implementation questions.

Are detectors configurable rules?
They are code-backed detectors with merchant baselines and evidence payloads. Teams can snooze or dismiss findings instead of tuning dozens of thresholds.
Where do alerts go?
Findings appear in the dashboard and can route to Slack based on severity, category, and merchant configuration.
Does schema drift count as anomaly detection?
Yes. Drift is tracked separately through taxonomy versions and drift rows, then surfaced alongside operational findings.
DEPLOY WITH EVIDENCE

Put anomaly detection into your live pipeline.

Start free trial ->

We use essential cookies to keep the site secure and functional. Analytics and third-party tags run only with your consent. See our Cookie Policy.

We use essential cookies to keep the site secure and functional. Analytics and third-party tags run only with your consent. See our Cookie Policy.