Per-execution firehose from Convex log streams. Call volume stacked by function, total error rate, and overall p95 latency over the window. A rising error line is bad.

Slowest functions by 95th-percentile execution time, and the daily database read/write bytes — the silent driver of the Convex bill. Watch for a function whose bandwidth climbs without a matching call-volume rise.

Green where every scheduled run succeeded that day, red where any failed. This is the view that was missing during the 23:20 scanTradeReminders incident.

Estimated Axiom ingest this month against the 500 GB free-tier cap, with an end-of-month projection. Estimated from event volume — Axiom doesn't expose true ingested bytes to a query token.

Un-redacted Convex exceptions from the dedicated honely-backend project. These carry the full server-side stack the client never sees.

Crash-free sessions and users, total error volume, the unresolved-issue queue. Events are flagged inverse — a rising line is bad.

JavaScript + server errors on the marketing site. Spikes here typically correlate with a recent deploy — check the unresolved issues for the root cause.

75th-percentile values for Core Web Vitals captured on the marketing site in the last 7d. Bands follow Google's thresholds — green / amber / red colour the p75 directly.