Skip to content

Stop tracking generated companion assets - #5

Open
everettVT wants to merge 1 commit into
mainfrom
codex/clean-generated-assets
Open

everettVT wants to merge 1 commit into
mainfrom
codex/clean-generated-assets

Conversation

@everettVT

Copy link
Copy Markdown
Collaborator

The companion previously required eight generated files to be committed: six task-review-*.js data bundles and two Cost vs Score PNGs. They added about 18.7 MB of duplicate content to the working tree. A fresh Pages build now generates them from the tracked record, audit metadata, task catalog, answer marks, and model registry. The generated outputs were checked byte for byte against the former committed files.

Removed in this PR

  • task-review-data.js and task-review-tasks.js: full and slim task catalogs generated from the pinned audit metadata and task catalog.
  • task-review-audit.js and task-review-published.js: browser wrappers around tracked audit JSON.
  • task-review-results.js: browser wrapper around results/rerun.json.
  • task-review-variants.js: derived box-scoring alternatives from tracked answer marks and the model registry.
  • results/dataset/cost-vs-score*.png: rendered from the committed record and registry.
  • Stale workflow assumptions and tests that read these files from the repository root. The Pages trigger now includes changes to the underlying build inputs.

Further cleanup opportunities, scoped separately

  • answers/*.json (~16 MB) are generated but are the only tracked per-question marks and responses used by Explore and the box-scoring review. Moving them out of Git requires a versioned external source plus a working Pages fetch/build path.
  • results/audit/gold_metadata/ and results/audit/gold-distributions.json contain the pinned question and reference material used to build the full task catalog. Removing them would require another reproducible, pinned source in CI.
  • results/rerun.json is derived but cannot be rebuilt from a clean checkout: the raw sweep records live under ignored results/runs/. Keep it until the record build has a complete tracked or published source.
  • results/archive/, scripts/archive/, and the audit paper snapshot are historical evidence rather than active site inputs. They can move to a release or research archive after their provenance links are reviewed.
  • original/ contains the released upstream code and visual assets, preserved with hashes; any size reduction requires a deliberate provenance policy and is outside this change.
  • docs/reviews/ and docs/plans/package-layout.md contain historical review and design notes. A documentation pass could mark superseded findings or move them to an archive.

Validation

  • Fresh uv run --group scripts vladbench build --steps site --site dist/pages succeeds.
  • All eight regenerated files are byte-identical to the versions removed from Git.
  • uv run ruff check ., uv run ty check, uv run pytest -q (138 passed, 11 skipped), and push hooks pass.

@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant