Official evaluation metrics hook for experimentation

Context

At KubeCon EU 2026, the experimentation discussion (recap) identified interest in having the SDK produce basic evaluation metrics that analytics vendors can consume for experimentation analysis.

Problem

Today, if you want evaluation metrics (e.g. how many evaluations per flag key, broken down by targeting reason, how many unique targeting keys have seen a given variant), you have to build that yourself. Multiple participants noted that this is a common need across experimentation platforms and the data is straightforward to collect.

Proposal

Define an official evaluation metrics hook that collects basic evaluation data:

• Evaluation count per flag key
• Evaluations broken down by reason (e.g. TARGETING_MATCH, DEFAULT, STATIC)
• Unique targeting keys per flag/variant
• Variant distribution per flag key

This hook could be published as an official OpenFeature contrib package. Analytics vendors (or an OFREP endpoint) could then consume this data without each needing to implement their own collection logic.

The hook approach is preferred over new API surface since this is buildable with the existing hook model.

Questions

• Should this be a spec recommendation or just an official contrib hook?
• What is the minimal set of metrics that provides value across vendors?
• Should this hook emit OTel metrics, or define its own output interface that vendors implement?

Related

• #370: Experimentation support: standardized context fields and experiment grouping
• #276: Implement Tracking in SDKs

Start by reading the existing hook model, then review related issues #370 and #276 for context on experimentation and tracking. Done requires agreement on the minimal evaluation metrics, whether this belongs in the specification or an official contrib package, and whether the output should use OpenTelemetry or a separate interface.