research-document RFR-006
Total verified-cost measurement
Total verified-cost measurement
Opportunity: Operationalize cost per verified useful outcome across compute, retries, latency, review, incidents, and recovery.
Unknowns: Which denominator predicts deployment value and how human/recovery costs change system rankings.
Origin and evidence: EV-E013; HY-E011; 12-research-and-engineering-roadmap.md North Star and Workstream D; 02-evaluation-specification.md Minimum Metadata.
Dependencies: RFR-002 and RFR-004; time and intervention logging.
Method: Prospective resource accounting, sensitivity analysis for uncertain cost weights, report task-level distributions and failed-attempt costs.
Outputs and success: Versioned cost schema, observed cost dataset, sensitivity dashboard/report. Success requires reconciliation of every run and rankings robust to plausible cost weights.
Recommended agent: Evaluation economist/operations researcher. Effort: medium. Expected gain: converts token-centric optimization into decision-relevant economics.