The v0.5.0-beta.1 fetch fallback did not help on macOS: the AU serves the UI from JUCE's custom juce:// scheme handler, where subresource loads work but fetch() does not. Promotes a generated models.js module - compiled into the binary and imported statically - ahead of the fetch tier, so the list arrives by the same mechanism that already delivers the page. loadModelList() order is now injected -> embedded -> fetch. models.js is generated from the committed ParametricEQData.cpp by generate_models_js.py, newline-joined and escaped to pure ASCII; verified byte-identical to the C++ names and node-tested across all four tier outcomes. Also moves the UI probe from ~1.5s to ~5s so a slow load is not misreported, adds an 'embedded' probe field, and puts the failure reason in the status bar so a screenshot is enough to diagnose.