ageval platform

Core owns configuration locking (lock), the Attempt phase pipeline, the EnvironmentProvider protocol, limits enforcement, trajectory archiving, and binding the independent evaluation result. The dataset owns the task interaction loop in run.py and grading criteria in evaluator.py.

Read order

1. docs/design/ — product and mechanism specifications (self-contained)
2. ARCHITECTURE.md — module layout and dependency direction
3. GitHub Issues — incremental delivery and acceptance tracking
4. Code / tests / examples — shipped runtime behavior

website/ is reader-facing documentation. When specifications conflict, docs/ and ARCHITECTURE.md take precedence; update the relevant documentation when wrapping up a phase or PR.

Ownership