[confiscate]

Summary

• recognize prompt-set markers in test UIDs that include evaluator suffixes
• keep official annotations excluded even for real -private test IDs
• cover shareable practice and official private-evaluator IDs in annotation export tests

Background

BenchmarkRun.compile_annotations() previously unpacked test UIDs into exactly five hyphen-separated fields. ModelBench CLI runs use the private evaluator, which appends -private; that caused parsing to fail and classified the prompt set as unknown. Non-shareable prompt metadata normally provided a second safeguard, but the explicit guarantee that official prompt sets never export annotations was not reliably enforced.

This is related to #1692 and hardens the existing annotation-export path added by #1700. It does not add SKU parsing or claim to implement the external SKU-to-CLI mapping.

Tests

• uv run pytest tests/modelbench_tests/test_run.py -q — 36 passed, 1 skipped
• uv run pytest tests/modelbench_tests -q — 411 passed, 1 skipped
• uv run black --check src/modelbench tests/modelbench_tests
• git diff --check

[github-actions]

MLCommons CLA bot:
Thank you very much for your submission; we really appreciate it. Before we can accept your contribution,
we ask that you sign the MLCommons CLA (Apache 2). Please submit your GitHub ID to our onboarding form to initiate
authorization. If you are from a MLCommons member organization, we will request that you be added to the CLA.
If you are not from a member organization, we will email you a CLA to sign. For any questions, please contact
support@mlcommons.org.
0 out of 1 committers have signed the MLCommons CLA.
❌ @confiscate
_{You can retrigger this bot by commenting recheck in this Pull Request}