[tillywoodfield]

What & why

Closes #21

A pipeline's transformations.md and changelog.md go onto its provenance, but a change to either left Dagster with nothing to track, so the graph kept serving the old text. Each export now takes a hash of those files as its code version, and its automation condition counts a new code version as an update. A docs-only change shows the export as unsynced, automation re-exports it, and the hand-run ingestion then shows stale, ready to run. Dagster does not pass staleness down a chain, so the ingestion turns stale once the export has re-run, which is one step later than the issue's wording.

Changes

• ExportSpec.documentation_version in data_commons/export.py hashes the DOCUMENTATION_FILES with a length prefix per file, so text moved from one file to the other gives a new hash.
• The nine exports built from an ExportSpec pass it as code_version. The dfo_shared, dfo_ontology and dfo_skills exports have no documentation files and are unchanged.
• EAGER_ON_DEPS_OR_CODE_CHANGE_WHEN_CHECKS_PASS in defs/automation.py is EAGER_WHEN_CHECKS_PASS with code_version_changed() added beside any_deps_updated() inside eager(), and the nine exports use it. A docs change that lands while a parent's blocking check fails waits until the check passes. EAGER_WHEN_CHECKS_PASS and the dbt models are unchanged.

Notes for reviewers

• After the first deploy all nine exports show "has a new code version", because earlier materializations recorded their run ID as the code version. code_version_changed() reads a first evaluation as unchanged, so automation leaves them alone and each flag clears when that export next runs.
• The code version covers the documentation files alone. An export's data version now derives from that code version and its inputs, so a hand re-export after a fix to export.py with unchanged inputs keeps the same data version and leaves the ingestion looking fresh.
• A missing documentation file fails the code location load, since the hash is computed when definitions load.

Checklist

• Addresses one issue, linked above
• Tests updated where relevant
• Docs updated where relevant
• Self-reviewed the diff

Testing: pre-commit, dg check defs, dbt build (1138 passed) and pytest (457 passed), run locally as in ci.yml. The new tests fail when the hash ignores a file or the length prefix, when an export drops its code version or condition, when the condition lacks the trigger, and when the trigger sits outside eager()'s since_last_handled().

🤖 Generated with Claude Code

[tillywoodfield]

@codex review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-27T16:48:54.786906Z	8a48a1f	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Bravo.

Reviewed commit: 8a48a1fd27

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".