Cloud9-Speaches 0.1.0 native five-OS candidates (UNSIGNED)

Cloud9-Speaches v0.1.0 — five-OS native acoustic candidate (UNSIGNED)

Source GitHub main merged PR #11 commit c925d52.

Native GitHub-hosted Windows x64, Linux x86_64/ARM64, macOS Apple Silicon and macOS Intel runners ALL PASSED:

• Real PyInstaller packaged FastAPI/Gradio and upstream CrispASR runtime starts outside source directory, /health and model/alignment API respond.
• Actually download, SHA256 verify and install the FastConformer French Q4_K GGUF, real 14.17 second French WAV and true measured 37/37 word CTC alignment without interpolation, CPU fallback on every platform.
• Windows NSIS installer / installed app startup / uninstall; native macOS app codesign ad-hoc validation and DMG creation; Linux CPU-only fallback archive <2 GiB.
• GitHub Actions lint + pytest green. Five OS artifact groups; 6 binaries with their 6 independently verified SHA256 manifests uploaded.

Production Cloud9 Speaches on Radeon 780M Vulkan separately passed real 37/37 acoustic alignment; CT300 AutoSubs and ScribeWatch pairing/health remains separately certified.

UNSIGNED / NOT FULL PRODUCTION CERTIFICATION: no Windows Authenticode certificate, no macOS Developer ID signature or notarization / Gatekeeper trust. Ad-hoc macOS codesign is not Developer ID. GPU Metal/CUDA on remote machines, physical microphones and full hardware-specific long-form benchmarks remain untested. Installers are prerelease candidates. Models are downloaded to user storage, not embedded.