Skip to content

hotato 1.8.1 — regression testing for voice agents

Choose a tag to compare

@quantumCF quantumCF released this 17 Jul 03:03

[1.8.1] - 2026-07-16

Fixed

  • State-verdict evidence binding. A state outcome verdict now binds to
    the specific evidence IDs it was judged on, and verify REFUSES (exit 2)
    when a bound evidence ID is missing from the record or does not cover the
    asserted state -- a verdict can no longer ride on evidence that was never
    about it.
  • File-based runs record origin: fixture. Runs scored from local files
    are recorded as fixture, never real; the serve views accept and label
    the fixture origin and keep the three origins (real / simulated /
    fixture) unmerged everywhere, so provenance survives from ingest to
    dashboard.
  • Release taglock toolchain detection. The build-exercising release tests
    now require a real importable build module (spec.origin is not None)
    instead of being fooled by a stray build/ namespace directory -- fixes the
    sdist-guard CI job, which failed on python -m build inside the extracted
    sdist tree.

Changed

  • README lead: "Regression testing for voice agents." The category line
    now leads every surface (README, PyPI, GitHub About); the wedge line sits
    second. How-it-works flowchart is top-down so GitHub's mermaid controls
    no longer overlap the verdict boxes; banner sized to 442 px.

Wheel and sdist are the exact bytes published to PyPI; verify the GitHub-signed build provenance with:

gh attestation verify hotato-1.8.1-py3-none-any.whl --repo attenlabs/hotato

SBOMs are CycloneDX, one whole-surface document plus one per dependency profile, generated offline from pyproject.toml and validated with scripts/gen_sbom.py --check.