fix(3.1.2): grandfather pre-3.0.43 datasets in dataset-size check

[FileSystemGuy]

Problem

A submitter re-validating against 3.0.46 is blocked on rule 3.1.2:

dataset size mismatch: actual files 4710038 < minimum required 4710039

Their dataset was generated with mlpstorage 3.0.24 (Jun 28), before the floor→ceil datagen alignment in f9d414d (first shipped in 3.0.43). Datagen back then sized datasets with integer floor division in rules/utils.py; the validator now re-derives the threshold with ceil, so it rejects a valid pre-alignment dataset by exactly one file. The submitter did nothing wrong — the official tool sized the dataset and reported it compliant.

Fix

Relax the rule 3.1.2 threshold from ceil to ceil - 1 for the v3.0 round (validator only; rules/utils.py datagen/runtime verifier stays on ceil).

Why ceil()-1 and not floor()

Datagen floored via integer // (exact), while this check recomputes the same formula as a float chain — so float drift can push floor() one above the recorded count and still reject the dataset we're grandfathering. ceil-1:

• equals floor() for non-integer ratios (no change in the common case),
• grants one file of slack exactly at the integer boundary where that drift bites, absorbing the bounded ≤1-file divergence,
• stays int-vs-int (no "N < N" display bug),
• still rejects genuinely undersized runs (short by >1 file).

Testing

All four CI suites pass (2938 + 890 + 238 + 228, zero failures). No assertion flipped, including test_true_undersized_submission_still_fails_3_1_2 (short by ~50k files, far beyond the 1-file relaxation).

Follow-up

Revert to plain math.ceil once the v3.0 round closes — post-3.0.43 datagen ceils, so newer datasets pass either way. Tracked inline in the code comment.

🤖 Generated with Claude Code

[github-actions]

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