Keep one current V2.1 dataset in the checkout

[MagMueller]

Main still carried two 55-task selector files and an obsolete 200-task snapshot alongside the current V2.1 dataset. Remove those three files and make the README point to one current 200-task V2.1 dataset, retaining the compatible BU_Bench_V2.enc filename.

Revision validation now reads its hash-pinned original baseline from Git history, with --base-artifact for archives or shallow clones. CI fetches the required history. The current dataset and encrypted judge review cases are byte-identical; V1/Stealth, scoring, runner settings and historical release assets are unchanged.

Validation: 68 unittest cases (65 passed, 3 optional skips); rubric validator confirms all 200 IDs, weights and canaries; Ruff, YAML parsing and diff checks pass. Verified the default loads all 200 tasks and V1/Stealth still load.

---

Summary by cubic

Keeps one current 200-task V2.1 dataset in the checkout. Removes the two 55-task selector files and the obsolete August 25 snapshot, and updates the README to reference only BU_Bench_V2.enc.

Refactors

• Revision validation reads its hash-pinned original baseline from Git history instead of the removed snapshot, verifying the SHA against rubric_revision.json.
• A new --base-artifact flag supplies the original encrypted baseline for source archives or shallow clones, with the same hash check.
• CI fetches the pinned baseline commit so validation runs there.
• The current dataset and encrypted judge review cases are byte-identical; V1/Stealth, scoring, runner settings, and historical release assets are unchanged. Normal benchmark runs need neither Git history nor the baseline.

^{Written for commit e5b79bd. Summary will update on new commits.}

[cubic-dev-ai]

All reported issues were addressed across 9 files

<sub>Heads up: you’re close to your flex budget. Increase your flex budget so reviews don’t pause.

Fix all with cubic | Re-trigger cubic</sub>