Skip to content

v0.8.0

Latest

Choose a tag to compare

@github-actions github-actions released this 13 Sep 13:53
· 8 commits to main since this release
7f98b72

Portable outcome reports and sample exchange

  • Ship the paired policy outcome explorer in the installed package and complete
    offline Stress Lab. Retain gains and losses, all seed groups and the full
    experiment denominator. Export and independently verify the complete report.

  • Include robot-reel-paired-outcomes.json as a checksummed release asset.
    Generate it with the installed wheel outside the checkout, reject an altered
    report, and compare the offline browser's filtered-view export with those
    exact bytes at desktop and mobile sizes.

  • Publish the descriptive 30-trial / 20-pair result tables on Hugging Face with
    a source manifest and a reproducible export script. The recorded experiment
    is unchanged; this release adds inspection and distribution tools.

  • Preserve Space homepage symbols and the Chinese link with HTML entities after
    public readback caught a UTF-8 character damaged at the hosting boundary.

  • Connect the Space and both READMEs to a curated Hugging Face model/data
    collection and the public feedback thread. Credit the exact recording sources.

  • Verify uploads anonymously against the tested CI artifact: every Git/LFS
    object plus the public manifest, homepage, three viewers and thumbnail.
    Bound CDN retries, reject stale or changed deployments and preserve unrelated
    remote files.

  • Host Cloth, Stress and Butterfly labs as a native Hugging Face static Space,
    with original recordings, a dedicated landing page and model/dataset credits.
    Publish the exact browser-tested artifact after the current main commit's
    checks pass; verify its full inventory and source identity before upload.

  • Make sample links visible and directly openable when embedded browsers deny
    clipboard access. Preserve the selected frame, condition and camera.

  • Open Cloth Lab sample JSON files, verify their source facts and restore their
    sample and camera offline. Reject changed or ambiguous files without moving
    the current view. Add cloth --verify-sample for independent standard-library
    checks against a verified source bundle.

  • Export Cloth Lab samples as 1920 × 1080 PNG figures with original mesh geometry,
    measured deformation, source clocks and a positions fingerprint, plus portable
    JSON records with full-precision metrics. Both work offline. Shared links now
    preserve camera rotation; failed PNG encoding can be retried.

  • Point the homepage, both READMEs and the published Stress viewer at the
    verified 0.7.1 downloads, installation guide and checksums.

Start with START-HERE.md. Extract either experiment ZIP and open its index.html
locally. Cloth Lab includes all 42,471 recorded vertex samples, original velocities,
the editable USD and native readback reports. The standalone cloth USD contains
the same bytes as the scene in the ZIP; import it into Blender at 30 fps.
For Stress Lab, import robot-reel-seed-09-review.json to inspect a selected moment
in the complete thirty-trial experiment. Both ZIPs are the exact artifacts
exported by the installed wheel and tested offline in Chromium at desktop and
mobile sizes. No GPU or Python is needed to use these recorded browser labs.

Stress Lab includes all four paired outcome groups and complete report export.
robot-reel-paired-outcomes.json was exported and verified by the installed CLI;
the offline browser produces the same report even after selecting a subset.
Use robot-reel stress stress-lab --paired-report robot-reel-paired-outcomes.json
to check it independently. Cloth Lab includes source-checked sample import,
portable JSON and 1080p figure export.

The wheel, source distribution and Rerun recording passed the same six validation
jobs. Open robot-reel-seed-09.rrd in Rerun 0.37.2. SHA256SUMS covers every asset.
The original recordings are unchanged; neither policy inference nor cloth
simulation was rerun for this release. PyPI publishing is configured separately.