Bellweather Audio Attest Is Public Source

It would have been easy to bury this work inside a broader testing suite and call it internal infrastructure. That would have made the system convenient, but less legible. Audio Attest deserved its own public shape because the idea is narrower than Bellweather as a whole and more specific than generic validation language.

The repo is about one thing: proving that rendered audio still matches a blessed baseline across the detector lanes Bellweather cares about. It does not host plugins. It does not load DAW sessions. It does not judge whether a mix is good. It compares rendered audio against a signed reference and makes that comparison inspectable.

That narrowness is the point. The more specific the claim, the easier it is to hold the release accountable.

The workflow starts with a known-good render. Audio Attest mints that render into a signed reference using a DSSE envelope and a VSA-style predicate over the detector baseline. Later, a candidate render can be checked against that reference with a public key.

That sounds more ceremonial than it is. The ceremony exists because audio regressions are slippery. If a baseline is just a loose file on disk, it is too easy for the reference and the candidate to drift together. If the reference is signed, the release process has a harder object to reason about.

This is not a claim that every audio question can be solved cryptographically. Taste cannot be signed. Musical judgment cannot be reduced to a hash. But the narrow question of whether a rendered output changed from a blessed reference can be made much more disciplined than a folder full of unnamed WAVs.

The boundary matters enough to say plainly. Audio Attest is not a replacement for a DAW, a plugin host, a product installer, a licensing system, or a release manager. It validates audio that something else already rendered.

That separation keeps the tool honest. Plugin hosting is its own problem. DAW session reproduction is its own problem. Product distribution is its own problem. Audio Attest does not pretend to own those layers. It owns the signed-reference proof around the rendered artifact.

That makes it useful in a release pipeline without turning it into the whole pipeline.

Barometer remains the right reference case because it sits at the overlap between the two public-source releases. Audio Core publishes the reusable metering code and the source-built Barometer reference stack. Audio Attest publishes a way to bless and verify rendered output from that world.

That does not make Barometer the product being downloaded from Audio Attest. It makes Barometer a concrete subject for proof. The same reason it helped Audio Core avoid being a diagram also helps Audio Attest avoid being a generic cryptography demo.

A proof tool needs an artifact to prove against. Barometer gives the public release a small, license-clean, source-built thing to point at without turning the repo into the Bellweather product platform.

A vague audio-regression claim is easy to make and hard to trust. Audio Attest keeps the checks explicit: loudness, true peak, loudness range, harmonic distortion, aliasing, stereo correlation and width, spectral masks, numeric hygiene, and related baseline comparisons.

None of those detectors can tell you whether a record feels finished. That is not their job. Their job is to make specific classes of drift visible when a render changes. If a nonlinear path starts aliasing differently, if loudness behavior moves, if stereo behavior shifts, or if a render starts producing invalid numeric output, the proof layer should have something concrete to say.

The value is not that the detector list sounds impressive. The value is that each lane names the kind of evidence it owns.

Audio Core and Audio Attest are deliberately different shapes. Core is a source library with a reference plugin. Attest is a headless proof layer for rendered output. One is about reusable audio code. The other is about signed evidence over what that code eventually produces.

Together, they are a more honest public story than either would be alone. Source without proof can become a pile of claims. Proof without source can become a black box. The pair gives Bellweather a public way to say: here is the code we are willing to show, and here is the proof discipline we are willing to show around rendered audio.

That is also why the two landing pages now mirror each other visually. They are not the same repo, and they should not pretend to be. But they should feel like two sides of the same release posture.

The docs can say how to run bpts bless and bpts verify. The landing page can give the shortest copyable path. The journal has a different job: explaining why this public surface exists at all.

The reason is not that the world needed another shorthand around audio testing. The reason is that Bellweather keeps coming back to the same pressure point: quality claims have to resolve into something inspectable. Sometimes that is source. Sometimes it is a test lane. Sometimes it is a signed reference. The public site should make those boundaries visible instead of smoothing them into marketing language.

Audio Attest is a small release in terms of surface area, but it carries a bigger editorial role. It says that proof is not just something Bellweather does privately before shipping. It is part of how the company wants to talk in public.