Debugging a failing delivery

Reading the timeline: what each status code usually means and what to do about it.

In short

Start from the attempt timeline: status, latency and the response body are all recorded.
A 401 usually means a rotated secret; a timeout usually means work done before acknowledging.
Endpoint health gives success rate and latency percentiles over the last day.

Read the timeline first

Each attempt records the HTTP status, the response time, an error class when there was no status, and the first kilobyte of the body. Most questions are answered before you open a log on the other side.

curl https://api.hookget.com/v1/events/msg_…/attempts -H "authorization: Bearer $HOOKGET_KEY"

What the usual answers mean

Symptom	Most common cause	What to do
401 or 403 from the destination	The consumer is verifying against an old secret	Check whether the secret was rotated and the consumer still holds the previous one
Signature mismatch reported by the consumer	The body was parsed and re-serialised before verification	Verify over the raw bytes
Timeouts	The consumer does its work before answering	Acknowledge first, process after
Intermittent 5xx	A deploy or a dependency of the consumer	Nothing — the retry schedule is for exactly this
Everything failing since a fixed moment	DNS, certificate expiry or a firewall change	Check the destination directly from outside your network
Nothing attempted at all	No active destination subscribes to that event type	Check the event-type filter on the destination

Endpoint health

curl https://api.hookget.com/v1/endpoints/ep_…/health -H "authorization: Bearer $HOOKGET_KEY"

# success rate over 24h, p50 and p95 latency, consecutive failures,
# and the reason if it has been disabled

An assistant can do the same reading: the MCP server exposes a diagnose tool that combines the URL check, the breaker state and the health figures into a verbal verdict.