Operations

Operational readiness

Most rollout issues are missing incident ids, exhausted credits, wrong keys, or silent degraded paths. Make these states visible before you ship.

Open reference Review billing boundary

Release criteria

Minimum requirements before the adapter should be considered production-ready

The moderation request alone is not enough. You need traceability, operator visibility, and explicit degraded-state behavior.

Include stable external idsEvery actor, message, asset, and scope id should be recoverable in your own logs and admin tooling.

Forward raw text and media inputsDo not collapse the event into your own moderation verdict before calling the platform.

Persist the incident idStore the returned incident id next to your local action so audit trails line up across systems.

Decide your degraded-state behavior explicitlyUnsafe content should not slip through because the moderation API became unavailable and nobody decided what happens next.

Support triage

When something looks wrong, inspect these surfaces first

Most rollout issues can be explained quickly if the operator can confirm key scope, incident presence, and remaining balance before diving into code.

API keysConfirm the adapter is using the intended workspace key and that its scopes still match the service.

Incident feedCheck whether the event was accepted, how it was classified, and whether credits were consumed.

Workspace creditsLow or exhausted balance will stop moderation work before enforcement ever reaches your adapter.

Diagnostic order for fastest triage

Check key scope, then incident ingestion, then credit balance, and only then investigate adapter code. This order avoids most false debugging starts.

Failure modes

Make silent failure difficult

The dangerous path is not a hard error. It is a system that silently stops moderating or loses the ids needed to explain what it did later.

Prefer visible rejection over invisible skip

Treat low-credit, wrong-key, or moderation-service failure states as first-class incidents in your product telemetry. Hidden degradation is what creates trust gaps during rollout.

Store enough metadata to reconstruct the decision

Incident ids, source event ids, scope ids, and any local enforcement outcome should be recorded together. Without that, moderation support becomes guesswork.