Require config.yaml beside point.yaml

[arav-agarwal2]

Summary

mlcommons/endpoints_policies#146 puts config.yaml ("Loadgen configuration file") back in §8.1's point directory, next to point.yaml. Since v1.0 the CLI had treated it as optional. This makes it required wherever the CLI reads a point:

Where	Change
runs create (runs/parser.py)	config.yaml joins the required run files; it must be a YAML mapping.
Bundle builder (submissions/builder.py)	An archive without config.yaml fails the build with a message citing §8.1. It was already copied into r<N>/ beside point.yaml, so the bundle layout is unchanged.
Checker	New rule config-file-present (§8.1), run on every r<N>/ whether or not its point.yaml loaded. Fails if the file is missing or isn't a YAML mapping. Nothing else in the checker reads it.
layout.py	CONFIG_YAML added to REQUIRED_RUN_FILES; the docstring tree shows it.

Also updated: the rule's text in messages.yaml, the README rule table and tree, and runs.md, submissions.md, endpoints-submission-cli.md and architecture.md. None of them call config.yaml optional any more.

The run-type fallback is unchanged: datasets[].type from config.yaml, then dataset_type from point.yaml. Only its wording changed, since "no config.yaml" can no longer happen.

Impact on existing data

• All 125 runs in published or in-review submissions already carry a config (checked read-only in production).
• All 101 point folders in test_submissions/ have config.yaml.
• All 9 point folders on NVIDIA's test review branch (submission/ad254a9c) have config.yaml, so they pass the new rule.

Testing

• pytest: 1720 passed, 1 failed. The failure is test_builder.py::…test_a_separate_accuracy_run_still_wins, which fails on main too.
• ruff and mypy src are clean.
• The five tests that asserted config.yaml was optional now assert the opposite.
• New tests cover a missing or non-mapping file in the parser, the builder and the checker, and that the builder ships config.yaml beside point.yaml.

🤖 Generated with Claude Code

[github-actions]

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