Skip to content

Effective observation review ​

The inbox keeps source reports and human review state separate. The effective review is a read-time view, not a replacement persisted report and not a causal evaluation of a skill.

Shared view ​

omk observe inbox --json includes effectiveExperienceReports, resolvedReviewSessions, and unappliedMetricAnnotations alongside the existing items array. GET /api/observe-inbox/view exposes the same three fields; both support skill filtering. The existing GET /api/observe-inbox item-list response is unchanged. Studio consumes the same domain projection.

Effective session indicators, basis codes, rule findings, inference, priorities, and skill totals belong to observability, not Studio's rendering layer. Deterministic priority uses the domain weighted score: scores of at least three mean review_first, positive scores below three mean sample_review, and zero means routine_sample. Domain reviewer findings can escalate this priority. Existing explicit LLM and manual-review precedence is retained, with provenance in resolvedReviewSessions.source.

Evidence and annotations ​

Canonical feedback attribution determines which feedback belongs to a skill, including permitted downstream feedback. Rejected feedback is excluded from the effective count. Where complete attributed event relations are available, other metric annotations are applied from that evidence rather than from the preview window.

A truncated or missing preview is not evidence of zero occurrences. When an annotation can be located but cannot be safely replayed from complete evidence, the stored count remains and the session's metric key is reported in unappliedMetricAnnotations. This is an evidence limitation, not confirmation that the annotation changed the total.

Raw reports, raw experienceReports, original trace evidence, and review state remain independently available in the inbox view model. Do not persist effectiveExperienceReports over those originals. Published measurement schemas, prompt bytes, and persisted statistical results are not changed by this projection.