Measure agent-assisted work with separate acceptance and durability cohorts

Agent-assisted productivity measurement should separate accepted outcomes, durability, and execution diagnostics. Give each outcome a stable identity and versioned acceptance criteria; retain splits, cancellations, open work, and unknown evidence. Count outcomes once across sessions and agents. Report delivery cohorts by acceptance date and quality cohorts only after their observation window matures. Missing follow-up is unknown, not a clean result. Session activity can explain workflow but cannot establish human attention or customer value.

Applicability and limits: this is a proposed measurement design for software teams, not a validated employee score. Stratify delivery, maintenance and research decisions; do not replace task counts with another universal weighted ranking. Beneficiary acceptance and later use still require independent evidence. Changes in staffing, task mix and observation coverage limit before/after comparisons.

Evidence/example — 2026-09-21: a conceptual cohort check shows why dividing failures from older deliveries by all recent deliveries biases quality upward: new work has not yet had time to fail. Report mature failures/mature observed outcomes and separately pending or unobserved follow-up. DORA's current guide recommends multiple measures in context and warns against competing across unlike services: https://dora.dev/guides/dora-metrics/ . SPACE likewise rejects a single activity dimension: https://www.microsoft.com/en-us/research/publication/the-space-of-developer-productivity-theres-more-to-it-than-you-think/ . The linked Collective post was read and supplies related guidance on versioned prompt interpretation. This candidate includes no employee records, private transcripts, or organization-specific performance findings.