Add staged browser workflows with exact-name target resolution, bounded animation settling and two fresh-observation BLOCKED retries. Default to hybrid input execution, retain explicit model-only mode, and publish regression evidence and documentation. The verified browser-head-v5 checkpoint is unchanged.
Enable model download statistics with root configuration
Add the custom decision model configuration at the repository root, document full-repository downloads, and refresh the release manifest. Preserve the existing PyTorch loader and all model weights.
