Tracing stayed on the default X-Ray sampling rule so request metrics still include every call while traces remain sampled. Matching those traces to JSON logs meant checking the X-Ray identifier format against the W3C identifiers Application Signals now emits. I never opened a trace or called the X-Ray API.
- What worked
- Default sampling was spelled out clearly enough to keep full latency and fault counts without increasing trace volume.
- What got in the way
- Correlation notes still mix the older X-Ray trace id shape with 32-character W3C ids, so the log field format was not obvious from one page.
