Preserve lifecycle evidence separately from evaluation projection

Reusable evaluator integration lesson:

A call can be correctly excluded from scoring while still exposing an evidence-plumbing defect. In the staging call evaluator, the canonical universal_context contained the node transcript but omitted lifecycle_summary/finalization, while sibling calls-table fields carried abrupt-end and provider facts. The evaluator therefore conservatively produced no pass/fail denominator, and the dashboard faithfully rendered an excluded result. Separately, the cohort classifier knew the names end_reason and telephony_provider but its PostgREST projection selected neither because the staging column mappings were unset, so the picker displayed unknown/unknown.

Recommendation: distinguish presentation correctness from evidence correctness; include canonical lifecycle/coverage facts inside universal_context, and ensure every classifier field is included in the source projection (or configure explicit disposition/source columns). Otherwise “not applicable” versus “not assessable” becomes inconsistent even when the top-level exclusion is defensible.

Observed: 2026-09-08T09:05:32.346Z Source provenance: e3-solutions/ender-inference evaluator-v2.1, staging evals-service, Launch 10 profile; repository code audit plus privacy-bounded staging smoke.