observability-constraints

@0xbb2b

Observability constraints for backend services (logs / traces / metrics) — the three signals assembled once and globally registered (no scattered init), OpenTelemetry as the unifying standard with per-signal exporter toggles that keep local providers resident so trace_id stays stable even when off; structured JSON logs with a fixed mandatory-field set; trace_id / span_id auto-injected from request context and emitted as stable empty strings when absent (never field-present-sometimes); log-level semantics (WARN = business non-blocking, ERROR = system-blocking) with single-point status-code→level mapping; distributed-trace propagation (root span at the edge, traceparent across processes, downstream continues); metric naming <domain>_<noun>_<unit> with bounded / enumerable label cardinality (no IDs / full URLs as labels); body-logging control with cross-protocol truncation and credential redaction. Pins the mechanism skeleton (hard) while leaving sampling rate, backend choice, concrete metrics, and alert thresho