An API health check can succeed while a particular carrier source is unavailable. Conversely, a fast cached lookup does not prove a fresh source request was fast. Reliability reporting needs to separate the kinds of work the service actually performed.
Trace’s reliability endpoint reports recorded lookup outcomes for your account. It is an operational view of your traffic, not a global carrier benchmark or a guarantee of event completeness.
Read the window and sample count together
GET /v1/analytics/reliability?days=7 accepts a 1–90 day window and separates outcomes by origin and carrier. Recording starts at deployment; historical observations are not invented and gaps are possible. Compare like-for-like windows before interpreting a change.
cURL · read-only account outcomes
curl "https://api.traceapi.dev/v1/analytics/reliability?days=7" \
-H "Authorization: Bearer $TRACE_API_KEY"Keep live and cached success separate
A cached success can help the customer experience during a busy period, but it should not conceal source acquisition failures. Stale fallbacks are older verified snapshots explicitly allowed by the caller. Failed results indicate the lookup did not produce an accepted result.
| Outcome | Useful interpretation |
|---|---|
| Live | A source acquisition returned usable evidence. |
| Cached | An eligible recent snapshot was reused. |
| Stale | A permitted older snapshot was returned after acquisition failure. |
| Failed | The request did not produce a successful tracking result. |
Interpret median and p95 with their sample sizes
Median duration describes a typical recorded request; p95 describes the slower portion of that observed sample. Neither is a maximum or an SLA. With very few requests, one slow lookup can change the percentile sharply.
Segment by carrier and origin before comparing latency. A monitoring worker, a direct request and a cached response can have different timing characteristics. Do not average a cached-heavy period and a live-heavy period and attribute the difference solely to an infrastructure change.
Connect a metric change to a concrete investigation
If failure share rises, inspect representative request IDs and whether the failures came from source acquisition, quota, input or the calling environment. Keep production customer payloads out of public issue reports. Test a small authorized parcel sample after correcting the cause.
For a migration or source change, capture the same outcome segments before and after, note the sample sizes and retain any missing observation periods. A dashboard that starts collecting data today cannot prove yesterday’s availability.
Common questions
Does a 100% success rate prove carrier data is complete?
No. Success means the lookup produced accepted evidence. It does not independently verify every carrier scan, accuracy or freshness beyond the returned metadata.
References & next steps
Implementation details were checked against Trace’s current API. External product details may change; verify the linked documentation before deciding.