The numbers behind the Dogma series
Every Dogma plug-in has a mathematical definition: the analog prototype of an EQ band, the static curve of a compressor, the ideal continuous-time version solved between the samples. We measure the built plug-in against that definition and publish the gap. The figures below are real — reproduced by our test suite on every build, measured on the shipping binary through its plug-in interface, not hand-picked.
The analog prototype or the exact continuous-time ideal — not another plug-in, and not our own engine checking itself.
We load the built .clap like a host and measure what actually ships, at 44.1 to 192 kHz.
The whole suite reruns in CI. A number that drifts fails the build, so what you read here is what you download.
Every band follows its analog prototype all the way to 20 kHz, with no latency. Where an ordinary digital EQ warps near the top of the spectrum, Dogma Q stays on the curve — measured across every sample rate.
Deviation from the analog prototype, by sample rate
Worst case across 17 EQ settings. The dashed line is 0.1 dB — roughly the smallest level change most people can hear. Every bar sits far below it.
The audio path is a single multiplication — no filter, no delay, no phase shift, only gain. The reference isn't "follows the static formula"; it's the ideal continuous-time compressor, solved between the samples. Dogma C measures itself against that.
What isn't taken comes out exactly as it went in. Between the S's the vocal is untouched — bit for bit — and the key filter is the exact analog response, not a bilinear approximation that drifts near the top.
Real spaces at zero latency. The engine is non-uniformly partitioned — a short head on the audio thread, a long tail in the background — so even a ten-second impulse response plays with no latency, and the result matches the reference convolution.
Measure any room, cabinet or piece of hardware and load it straight into IR-REV. The capture chain uses regularized spectral division — and against a known reference sweep it beats the standard Farina and Novak methods, with a full ISO 3382 report on every measurement.
Reconstruction error vs the known reference IR — capture method
RMS magnitude error on the ideal reference sweep, lower is better. Our regularized division reaches less than half the error of the published Farina and Novak methods.
These are a selection of the headline figures. Each plug-in's full report — hundreds of measurements, at every sample rate, on the shipping binary — is produced by our test suite and travels with the build. Want to run your own tests? The beta is open.
