Feature - Elo rating estimation & continuous evaluation pipeline

[Codex-Crusader]

Summary

Create a pipeline that evaluates checkpoints by playing them against previous bests and estimates an Elo (or TrueSkill) score over time.

Motivation

Automated rating helps track progress and decide when to promote a checkpoint.

Acceptance criteria / Definition of Done

• Implement an evaluation runner that:
  • Runs N matches between a new checkpoint and current best with configurable time controls.
  • Collects match outcomes (win/draw/loss).
• Implement a simple Elo/TrueSkill estimator that converts match results into a rating or rating delta.
• Produce a plot/report (CSV + PNG) showing rating over time for checkpoints.
• Document the evaluation threshold for model replacement (configurable).

Implementation notes

• Use existing libraries (e.g., trueskill) for rating if desired.
• Keep evaluation reproducible (fixed seeds/time controls) and make it runnable in CI for small sample tests (but full evaluation should run offline).

Complexity

Medium–High — requires evaluation harness + metrics code.