v0.3.0: Initial public prototype

The first public release of Scene Acceptance, an experimental harness for evaluating saved OpenUSD content against application-owned requirements.

Includes five built-in pack families, an independently installable mesh-budget example, architecture diagrams, setup instructions and retained evidence. The evaluator produces findings; the calling application owns producer retries and release.

The runtime is the same tested 0.3.0 implementation as the earlier article companion. This release adds standalone project documentation, licensing and automated replay. The archive includes source, fixtures, historical results and a file-hash manifest.

Validation from a fresh archive extraction and packaged installation on Python 3.12 / macOS arm64:

• 195 software tests passed.
• 22 pack cases and 32 mesh cases retained their expected decisions and check statuses.
• Four content probes, three physics-configuration cases and four CPU simulations reproduced the retained observations.
• Both README quickstart commands produced the expected acceptance/rejection.
• All 796 manifest-listed files verified before and after replay. Counts overlap; no new model calls were made.

The first pack run's three false acceptances remain available beside the corrected results. Historical model outputs are separate from constructed faults. Texture decoding and the restricted MuJoCo behavior runner are separate experiments, not installed runtime packs. No physical validation, robot policy performance or comparative productivity is established.

Download scene-acceptance-0.3.0.zip for the complete project. scene-acceptance-SHA256SUMS.txt verifies that archive. verification.json records the fresh replay; README and docs/EVIDENCE.md explain how to repeat it. The GitHub-generated source archives contain the same tagged source but have different archive hashes.