[ayam04]

What changed

to_grid guarded the rate with if rate_hz <= 0.0. nan compares false against every comparison, and inf is positive, so both got past that check and reached Fraction(rate_hz) in _grid_timestamps_ns, where they surfaced as ValueError: cannot convert NaN to integer ratio and OverflowError: cannot convert Infinity to integer ratio instead of a rate error.

The rate now goes through require_finite_float from hflow._field_guards before anything else, so a non-finite rate is refused with rate_hz must be finite, got nan. The existing non-positive branch and its message are unchanged, which keeps rate_hz must be positive, got 0.0 for 0 and negative rates. The shared guard also rejects bool, which previously passed as 1 Hz.

I kept require_finite_float alongside the existing branch rather than collapsing both into require_positive_float so the message already documented for 0 and negative rates does not change. Say the word if you would rather have the uniform must be > 0 wording from the shared guard; it is a one-line change.

Tests

test_to_grid_requires_a_finite_positive_rate is parametrized over 0.0, -1.0, nan, inf and -inf, asserting the message each case has to produce. It replaces the single-case test_to_grid_rejects_nonpositive_rate.

Validation

• pytest tests/test_resample_derive.py -q -> 23 passed
• RED proof: with the new guard line removed, the nan, inf and -inf rows fail (3 failed, 2 passed) with cannot convert NaN to integer ratio / OverflowError / the non-positive message; restored, the test passes (5 passed)
• ruff check + ruff format --check on both files: clean
• ty check: clean

Fixes #726

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[greptile-apps]

Confidence Score: 5/5

[Low impact] This PR appears safe to merge.

Summary

to_grid now rejects non-finite rates before building timestamps.

• to_grid now rejects non-finite and boolean rates before building a grid.

_{Reviews (1) · Last reviewed commit: "fix(resample): reject a non-finite to_gr..." · Reviewed by Greptile}

[ayam04]

Full suite provenance, since it is not all green locally:

• pytest -q on this branch: 2506 passed, 41 failed, 8 skipped (387s)
• 36 of the 41 are the packaging tests (test_packaging.py 17, test_packaging_manifest_parsing.py 19), which cannot build the native overlay in this WSL image.
• The other 5 (tests/test_cli.py x3, test_server_catalog_tables.py, test_server_episode_dossier.py) fail identically on a clean origin/main worktree (85178e7) with no changes applied: 5 failed, 25 passed. So none of the 41 are from this diff.

Nothing in the failure set touches resample, to_grid or the rate guard.

[kstonekuan]

Closing as superseded by #728, which merged the non-finite rate fix using the shared positive-float guard.