Completes the topology-driven metrics feature from PR 135 by implementing the four follow-up issues, one commit each.

Scenario overrides for metric values (closes #136)

Scenarios can override a metric's value distribution during their window, at operation scope (gateway.handle) or service scope (bare gateway, metrics only):

scenarios:
  - name: cpu spike
    at: +2m
    duration: 5m
    override:
      gateway:
        metrics:
          gateway.cpu.utilisation:
            value: 0.95 +/- 0.02

The engine pushes resolved overrides to observers each cycle via a new OverrideObserver interface; MetricObserver applies them when sampling values for counters, updowncounters, histograms, and gauge callbacks. Overlapping scenarios merge per metric name with the existing priority semantics. Validation rejects overrides of span-derived metrics, metrics not defined at the scope, and non-metric fields on service-scope keys.

Random-walk gauge values (closes #139)

Gauges gain an optional walk: mean-reversion timescale that replaces independent per-collection samples with an Ornstein-Uhlenbeck process — consecutive samples are temporally correlated while the stationary distribution still matches the configured mean and standard deviation. Optional min/max bounds clamp gauge values:

- name: gateway.cpu.utilisation
  type: gauge
  value: 0.65 +/- 0.1
  walk: 30s
  min: 0
  max: 1

White noise remains the default; walk timescales relate to wall-clock time between collections.

Rate-driven independent metrics (closes #137)

Topology-defined counters, updowncounters, and histograms accept an interval: field that emits sampled values on a wall-clock ticker instead of per span, so a service with no traffic still produces metric data:

- name: gateway.gc.pause.duration
  type: histogram
  unit: ms
  value: 2.5 +/- 0.8
  interval: 10s

MetricObserver.Start() runs one emitter goroutine per interval instrument and stops cleanly before meter provider shutdown. Interval emitters honour scenario value overrides. Gauges (already collection-driven) and span-derived metrics reject interval at validation.

Semconv-aware metric validation (closes #138)

motel validate now checks metric names against the vendored semantic convention registry. Where a name matches a known semconv metric, mismatched instrument types and units are reported as warnings on stderr — never errors, since users may intentionally deviate:

warning: service "gateway": metric "http.server.request.duration": unit "ms" does not match semantic convention unit "s"

Adds Registry.MetricByName to pkg/semconv. This caught non-compliant units in our own docs examples, which are updated to unit: s / unit: "{request}". The semconv attribute shorthand extension mentioned in the issue is left for a separate change.

Testing

• 28 new test cases across config validation, scenario merging, observer behaviour (override application, walk correlation and mean reversion, bounds clamping, interval emission and stop), the semconv registry, and the validate command
• Verified end to end: a topology combining a walking bounded gauge, an interval histogram, and a scenario override produces the expected values via motel run --signals metrics --stdout
• make test and make lint pass

https://claude.ai/code/session_01TgWavXRfzQiTWDZccuS2pf

Generated by Claude Code