JigEQ ABC implementation report
Task list
The plan in docs/jigeq-abc-plan.md is a good benchmark goal, but the practical requirement is a reusable measurement harness that can judge any equalizer without depending on one UI. The repo already had the surrounding checks for plugin validity and render budget; the missing piece was a direct, repeatable response check for tone gain and frequency sweeps.
Completed
- Added
src/host/PluginAnalysis.jswith:measureToneGain(), which renders a single sine through a plugin and reports input RMS, output RMS and gain in dB.measureFrequencyResponse(), which sweeps a list of frequencies and returns a response curve.- shared helpers for
rms()andgainDb()so the same logic can be reused by a UI benchmark later.
- Wired the CLI in
bin/check-plugin.jsto accept--frequencyand--sweepso an equalizer can be checked from the command line without writing a bespoke harness. - Added
tests/host/PluginAnalysis.test.jsto prove the measurement matches a pass-through EQ in its default state.
Completed
- Build the three interface variants from the same DSP backend: A = minimal UI, B = SVG skin, C = HTML UI.
Remaining for the full ABC set
- Add a fixture that logs all three response curves into one table so the UI trade-offs are visible in the same way as the engine trade-offs.
- Measure startup time, CPU budget and visible latency for each implementation.
- Capture the lessons learned into a final comparison report, keyed to the exact DSP and UI decisions that shaped the result.
Visual harness
npm run jigeq:harness starts the local server on port 8750 and opens
/jigeq-abc.html. The page uses the normal profile validator, verified plugin
loader, generated panel, SVG skin renderer and sandboxed HTML editor. Its
stereo white-noise source feeds the selected plugin and reaches the browser's
default audio output after the output-level control. Press Start white noise
in the page to grant audio playback; only the selected EQ is routed at a time.
The harness compares the three interfaces visually and audibly. It does not measure plugin CPU use or render latency.
Practical conclusion
The main work needed in practice was not a second equalizer engine but a reliable way to measure the one engine in many UI shells. The response helpers now give the project that measurement path, which is the part a real UI benchmark must share across A, B and C.