Six independent detectors live in workers/cron, reading from your event log, computing baselines, and writing graded findings to the insights table. Each finding carries severity (critical → info), category, evidence (jsonb), and an action_type/action_payload pair the dashboard renders as a one-click fix.
You don't configure thresholds. You don't write rules. The detectors derive baselines from your own 7-day history and flag deltas that matter. Open / dismissed / snoozed / fixed — your team works the queue like Sentry, not like a settings page.
An IT contractor rotates the Meta system user token. Your CAPI dashboard says "events firing" — and they are. They're just being rejected at the edge with err.190. Match keys vanish. Two weeks later your CMO asks why CPM is up 40%.
Your dev team renames customer_email to user.email in the new checkout. Identity match-rate drops 30 points overnight. Meta thinks 41% of your purchases are anonymous. Optimisation degrades silently for the next two ad cycles.
A flash sale doubles your traffic. By day 22 of the billing cycle you've already burned your monthly event budget. A threshold alarm fires at 100% — too late. A trajectory model flags the trend at day 19, when you can still upgrade calmly.
Watches per-platform delivery success vs 24h baseline. Flags when Meta CAPI / TikTok / GA4 / Klaviyo drop ≥5pts.
Cross-references the 32-event taxonomy with what's actually firing. Catches missing add_payment_info, view_item_list, search.
Detects same event_id firing within 60s (race conditions, double-tag installs, retry storms post-deploy).
Tracks % of events with usable match keys (email, phone, fbp, fbc, external_id). Flags regressions vs 7d baseline.
Linear-projects monthly event volume against tier cap. Fires 3-7 days before overage at 80% / 95% / 100%.
p95 latency per platform per 6h window vs rolling baseline. Catches Meta API degradations, DNS issues, regional drift.
{
"platform": "meta_capi",
"success_rate": 0.873,
"baseline": 0.961,
"retry_count": 1832,
"error_taxonomy": {
"OAuthException.190": 1804,
"RateLimit.4": 28
},
"window": "24h",
"events_affected": 14211
}Meta system user token rotated by client without telling agency. CAPI delivery on northfield.co dropped from 96% to 31% at 02:47 UTC. Daily detector flagged at 09:00 with error taxonomy 'OAuthException · 190 · access token expired'.
Account team had a fix-it ticket open before the client noticed missing conversions in Ads Manager. Recovered ~6h of match keys via TrackLayer's 24h replay buffer.
Migrated checkout from Shopify to a custom stack. Match-rate baseline 78% (email + fbp + fbc). Day 2 post-launch: detector caught regression to 41% — fbc cookie not being read on the new domain due to subdomain mismatch.
Caught 28 hours after deploy, before the next Meta optimization window. Without the detector, EMQ would have silently dropped for 2-3 weeks before someone noticed CPM creep.
Halloween campaign drove unusual spike. Day 22 of billing period: 2.7M of 3M Starter quota burned. Detector projected 3.6M EOM with 8 days remaining — flagged 3 days before any other tool would.
Operator upgraded to Scale ($79 → $249) the same day. Avoided ~$30 soft overage and the panic of seeing it on the next invoice. Linear projection beats threshold alarms when traffic is non-uniform.
Insights detectors run for every paid tier — there is no extra add-on. The upgrade ladder is where findings reach you and how often the detectors run.
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.