This accompanies the process article and playable collection. It separates the October 1–2 engine-development record from the October 2 publication checks.
Development record
Claude’s groovebox came from Ferenc PR #1, pinned locally at b9d5bb0e1cbc5d07484a7c288f4336879b822c15. The PR describes its browser smoke test and explicitly leaves physical audio/MIDI checks open. The local integration later brought its parameter catalog into the YAML specification.
The local engine verification report reached 99 passing Node tests, plus PHP validation/SQLite/filesystem checks and generated-spec parity. It records an actual Chrome AudioWorklet measurement at 48 kHz: an A4 sine measured 439.985 Hz, RMS 0.01817, and peak 0 after Panic. Imported WAV persistence, generated-stream capture/Freeze, and a 390 × 844 mobile overflow check were also recorded. A generated stream is not a physical microphone test.
The local progress ledger and regression tests document concrete failures: pending transport starts after Stop, note ownership collisions, effects refilling after Panic, small rounded knob changes, browser worklet structuredClone, detached decoding buffers, and short-cycle granular region wrapping. Their fixes continued into October 2. These records support a history of iteration, not a controlled measurement of the rules’ contribution.
Public release checks, October 2
The publication adds a static export under /fun-and-games/instruments/, a replaceable browser project store for Astrax, and shared original demo loading for Claude’s groovebox. The exported Astrax specification includes the public deployment contract and version mounts.
| Check | Result |
|---|---|
| Complete Node suites | 104 tests passed, zero failures |
| PHP validation, SQLite, filesystem | Passed; existing local server behavior retained |
| YAML/generated catalogs | Compiler parity passed |
| Static export | All three nested entries, links, module imports, and worklets resolved |
| Original bank | Exactly six WAVs; content hashes, PCM format, DC, peaks, boundaries, and repeat generation checked |
| Astrax, actual Chrome | Original cycle loaded; sequence produced nonzero output; Panic peak 0; saved project restored after reload |
| Claude groovebox, actual Chrome | Factory sequence and original cycle produced nonzero output; all six public slots loaded; autosave restored after reload; hard note cleanup gave dry peak 0 |
| Fischwerke, actual Chrome | Original texture loaded in Grain; phrase produced nonzero output with active grains; Panic peak 0; saved Grain patch restored after reload |
| Browser console | No error or warning messages during those playback/persistence checks |
| Screenshots | Captured from the running exported apps; P-6 image is a static design drawing |
Digital output was inspected with Web Audio analysers. That verifies a functioning browser signal path; it does not establish acoustic quality, device latency, or physical MIDI timing. After upload, all three instruments produced nonzero output at their live HTTPS URLs, the six originals loaded, and their consoles showed no errors or warnings. All 101 published files returned HTTP 200 and matched the reviewed build byte for byte. All four article images loaded in the live page.
Samples and design sources
The original bank download includes reproducible standard-library Python synthesis, seed 20261002, an independent checker, and measurements. Its provenance manifest identifies each WAV by its SHA-256 prefix. No existing source audio is read by the generator. The six sounds and generator carry CC0-1.0.
The P-6 drawing is a proposed MIDI control surface. Roland’s documentation supports discrete Auto Pan modes and Filter Envelope Depth from 0 to 255. Those display ranges alone do not establish CC conversion functions. The Web Audio specification describes input-buffer detachment during decoding.
Physical MIDI, real microphone/interface capture, acoustic quality, latency, long sessions, and a working P-6 remote remain outside this evidence. Wider Fischwerke scenes/song/autosampler workflows remain designs.
Evidence levels: implementation [4 HIGH | local specification, contract/regression tests, verification and progress records]; public browser behavior [3 MODERATE | this release’s Chrome checks]; hardware behavior [1 SPECULATIVE | no physical-device verification].