Skip to content

feat: add clean-desktop mode for deterministic WAA parity#106

Merged
abrichr merged 1 commit into
mainfrom
chore/evals-main-clean-review-20260304
Mar 4, 2026
Merged

feat: add clean-desktop mode for deterministic WAA parity#106
abrichr merged 1 commit into
mainfrom
chore/evals-main-clean-review-20260304

Conversation

@abrichr
Copy link
Copy Markdown
Member

@abrichr abrichr commented Mar 4, 2026

Summary

  • add a deterministic desktop policy in WAALiveAdapter to suppress OneDrive/toast/popover UI noise during task resets
  • add tray parity controls to keep network/audio surfaces visible (HideSCANetwork=0, HideSCAVolume=0, EnableAutoTray=0)
  • add CLI flags --clean-desktop, --force-tray-icons, and --waa-image-version to run, live, and smoke-live
  • thread the same flags through scripts/run_dc_eval.py and scripts/run_eval_pipeline.py
  • persist run environment metadata (requested + observed profile) into each run's metadata.json
  • document deterministic mode usage in README and script examples

Why

External eval users reported OneDrive popups and unstable tray UI (network/audio) causing click-coordinate failures and train/eval mismatch. This adds a first-class parity mode and run-level provenance.

Validation

  • python3 -m py_compile on touched Python files
  • attempted uv run ruff / uv run pytest but local env is blocked by numpy build failures on arm/py3.13 (NEON intrinsics not available with soft-float ABI)

@abrichr abrichr merged commit e26fabe into main Mar 4, 2026
1 check passed
@abrichr abrichr deleted the chore/evals-main-clean-review-20260304 branch March 4, 2026 20:21
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant