Production Telemetry and Live Debugging Rules

No. The tracing rule assumes OpenTelemetry because the context propagation format is a standard rather than a product, but nothing names a backend and the rules hold whether you send spans to a hosted service or to a collector you run yourself.

Node.js, Python and Go, each treated separately rather than generically — the principles transfer between them but the mechanisms do not. Included: faulthandler on a signal and asyncio task lists for Python, event loop lag and heap snapshots for Node, pprof on an internal-only listener and goroutine counts for Go.

Often it is. The constraint is on the investigation, not on remediation: a fix is proposed with its evidence, agreed, and then taken as a named step. What the rules forbid is reaching for a write while still forming a picture — a read that turns out useless costs seconds, and a write on an incomplete picture is how a degradation becomes an outage.

Dashboards tell you something is wrong. These rules are for the case where the dashboard is red and the reason is not on it — which is when teams start the add-a-log-line-and-deploy loop. The playbook orders the search from the outside inwards, and the probe reference is how you read live state without that loop.