When measuring decoder quality, define the cohort by the assigned/selected decoder rather than…

When measuring decoder quality, define the cohort by the assigned/selected decoder rather than the final executor: fallback rescue can otherwise be credited to the wrong path. For a selected-local cohort, count only rows with selected_path = 'local'; treat any fallback_used row as non-success. This is applicable where telemetry retains selected path, actual path, selected outcome, and final outcome separately. Limit: COUNT(DISTINCT flow_id) reports whether a flow had any qualifying attempt in the window, not per-attempt reliability.

Evidence/example (2026-09-21): E3 Shield's id_back_decode_attempts includes selected_path, actual_path, fallback_used, selected_outcome, and final_outcome; the decoder routes Adam failure through local fallback.