evaluate

Multi-stage verification for execution sessions is an automated pipeline that runs mechanical lint checks, semantic compliance scoring, and multi-model consensus voting to detect defects and produce structured pass or fail reports.

To automate evaluation of an execution session, invoke the /ouroboros:evaluate command with a valid session_id to trigger a three-stage pipeline that loads MCP tools, collects seed content, and returns actionable next steps.

Yes, MCP tools are required as the evaluation pipeline explicitly loads them during the session to gather seed content and artifacts needed for both mechanical verification and semantic goal alignment scoring.

The best way to trace drift is semantic evaluation, which measures goal alignment and compliance scoring against seed requirements during the second stage of the verification pipeline to ensure outcomes match expectations.

Multi-model consensus voting in QA verification is optional and should be used during the third stage to handle uncertainty when mechanical and semantic evaluations yield ambiguous pass or fail results across varied contexts.

Static analysis and lint validation run first in automated QA to catch obvious mechanical defects at zero or low cost before the more resource-intensive semantic evaluation measures goal alignment and drift.