Prediction-vs-reality capture per action

[CodeBuildder]

Context

Compares experiment predicted_outcome to actual post-experiment system state. Parent: #42.

Acceptance Criteria

• After each experiment, recorder reads the prediction event and post-experiment findings from the bus
• Actual outcome determined from metric snapshot within configurable window after experiment end
• Calibration record written to the shared memory store calibration endpoint
• Integration test: run experiment with known outcome → calibration record created with correct brier_score

Dependencies

Blocked by: none | Blocks: none

Notes

No model tier.