[ Competency domains ]
FDE-05 · Evaluation, verification, and reliability
Convert the mission contract into layered, repeatable evidence that detects behavior regressions across outputs, trajectories, system outcomes, and operating conditions.
- Elaboration
- Builds representative cases from real workflow patterns, including edge, adversarial, and abstention cases. Combines deterministic checks, model-based graders, human review, and production signals according to consequence. Inspects traces and failure clusters before changing prompts, models, tools, or architecture. Uses fixed thresholds, regression canaries, and explicit HOLD or SHIP decisions.
- Source record
- Forward-Deployed Engineering Standard FDE-S/1.0, competency FDE-05
[ Related terms ]
[ Where this is discussed ]
- Build the third layer before anyone hires you for it.FIELD GUIDE · Sep 5, 2026
A definition here is not a claim about any person, engagement, or outcome.