Stop treating pipeline-assigned fields as system_desc.json fields

[arav-agarwal2]

Part of mlcommons/endpoints_policies#143. Rules side: mlcommons/endpoints_policies#144.

The pipeline assigns the submission ID (rules §8.5) and the submission and publication dates, and the bundle records the ID only as the <submission_id>/ directory (§8.1). The CLI still treated all three (submission_id, submission_date, publish_date) as system_desc.json fields, left over from v0.7:

• Every bundled system_desc.json got all three as null. SystemDescription declared them, and the builder writes model_dump(), so the keys were added even when the submitter's file had none. (The submission_id entry is the one reported in the issue.)
• A v0.7 numeric value failed system-description-valid, because the fields were typed str | None.
• Points whose copies differ failed system-description-consistency, since the fields were compared like any system field.

Changes

• New PIPELINE_ASSIGNED_FIELDS in models/file/system.py names the three keys. SystemDescription no longer declares them; extra="allow" still parses a leftover key of any type, and model_dump no longer invents them.
• The builder drops them when loading a run's system_desc.json, next to where it already normalises division and the availability spellings.
• The checker leaves them out of the consistency comparison (_IGNORED_SYSTEM_DESC_FIELDS), for bundles that reach it without going through the builder.

Nothing read these attributes on SystemDescription.

Testing

• New tests in test_models.py, test_checker.py and test_builder.py, each run for all three fields. 21 of the 27 new cases fail on main (the other 6 pin that string and null values keep parsing); all pass here.
• pytest: 1609 passed. ruff check src, ruff format --check on the changed files, and mypy are clean, and the Sphinx docs build has no new warnings.

🤖 Generated with Claude Code

https://claude.ai/code/session_01N2grVghMMtp7LPurYJ5o35

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[Copilot]

🟢 Approval recommended

The implementation consistently addresses legacy parsing, generated output, and cross-point comparison with comprehensive targeted tests.

0 open findings

What changed in this PR

Removes pipeline-assigned metadata from generated and validated system_desc.json files while preserving compatibility with legacy submissions.

Changes:

• Defines and removes pipeline-assigned fields during bundle construction.
• Excludes legacy values from consistency checks.
• Adds model, checker, and builder coverage for all affected fields and value types.

File	Description
src/​submission_checker/​models/​file/​system.py	Removes declared pipeline fields and defines their shared names.
src/​submission_checker/​checker.py	Ignores pipeline fields during consistency checks.
src/​endpoints_submission_cli/​submissions/​builder.py	Drops stale pipeline fields before validation and output.
tests/​submission_checker/​test_models.py	Tests parsing and serialization behavior.
tests/​submission_checker/​test_checker.py	Tests validation and consistency behavior.
tests/​endpoints_submission_cli/​submissions/​test_builder.py	Tests omission from generated bundles.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.