[Hydrodynamical]

The README pointed to archived entry points, the offline mock never reached a scored prediction, and sweep resumption could conflate prompt and sampling settings. This change gives visitors a tested Python 3.12 quick start, explains the experimental-design task, and records the limits of the available historical evidence.

Changes:

• Pin the numerical environment and optional provider dependencies; add an offline GitHub Actions workflow.
• Update parser tests to the active protocol, repair the difficulty-aware mock, and label mock logs correctly.
• Preserve prompt, temperature, and reasoning settings in sweep identities and logs; prevent raw-log filename collisions.
• Replace stale paths, output descriptions, budget claims, and the unsupported single-seed comparison in the README.
• Reconcile manuscript protocol/evidence descriptions and rebuild the PDF, preserving numerical result tables.

Validation:

• 47 offline tests pass on Python 3.12.12, covering all four mock difficulties and configuration-specific retries/merges.
• The documented mock command writes a scored JSON log and dashboard.
• Optional API dependency pins resolve; no live provider benchmark was run.
• Manuscript compiles, affected pages were visually checked, and no new layout warnings were introduced.

The complete source-run archive behind the historical tables and figures is still absent. The README and manuscript now state that limitation explicitly.