Neuroimaging with Sounio

Executive reading

This route exists to make one claim legible: diagnostic software should be able to preserve uncertainty all the way from signal handling to the final threshold, and sometimes the right answer is refusal instead of classification.

Neuroimaging is a brutal domain for soft language claims. The signal is noisy, the biological story is partial, and the temptation to turn fragile evidence into a decisive-looking output is always there. That makes it a useful pressure test for Sounio.

If the language cannot carry ambiguity honestly through this kind of pipeline, then its epistemic posture is not worth much.

What this route is trying to prove

Sounio should be able to support diagnostic honesty, not only numerical convenience.

In practice that means:

• signal-processing stages should not erase the confidence story
• thresholds and downstream decisions should be able to see degraded evidence
• the pipeline should have room to refuse a strong conclusion when the surface is too weak
• the language should help keep provenance and uncertainty visible instead of hiding them in comments, notebooks, or post-hoc dashboards

That is what “diagnosis” means on the homepage. It does not mean “medical AI heroics.” It means keeping the software honest about what it does and does not know.

A skeptical reader should ask

• Is this route actually about neuroimaging, or is neuroimaging just set dressing?
• Are there concrete signal-handling surfaces here, or only abstract prose about trust?
• Does the pipeline have a refusal posture, or only a confidence annotation?
• Is the claim clinical deployment, or language pressure from a high-consequence domain?

The correct answer is the last one. This route is on the site because it forces the language to confront diagnostic ambiguity in a domain where false certainty is expensive.

HRF testing

fn test_hrf() -> bool {
    let params = hrf_params_canonical()

    // HRF should peak around 5-6 seconds
    let h5 = hrf_value(5.0, params)
    let h6 = hrf_value(6.0, params)
    let h10 = hrf_value(10.0, params)

    h5 > h10 && h6 > h10
}

The example is intentionally foundational. The route is not trying to fake a full diagnostic product. It is showing the kind of signal-processing substrate that becomes meaningful once the language can preserve uncertainty, provenance, and explicit thresholds through the rest of the pipeline.

Why this matters to the language

Diagnostic domains punish cheerful abstraction.

It is easy to market a language as expressive. It is harder to show that the same language can remain disciplined when the evidence is ambiguous, the signal quality is unstable, and downstream readers want a definitive answer anyway.

That is why neuroimaging belongs in the flagship set. It demonstrates pressure from a domain where “just give me a number” is often the wrong cultural default. Sounio should be shaped by exactly those domains.

What would convince a skeptical reader

• clear examples of signal handling that remain tied to uncertainty-bearing downstream logic
• explicit discussion of when a pipeline should refuse to classify or escalate for review
• honest separation between “research pressure” and “clinical deployment”
• a route back to the artifacts, libraries, or proofs that actually ground the claim

Without those elements, this page would just be neuroscience-flavored branding.

Where the boundary still is

This route does not claim:

• that Sounio is already a deployed clinical neuroimaging platform
• that the current examples solve diagnostic inference end to end
• that every statistical or signal-processing uncertainty has already been modeled in the public surface

The claim is narrower and more serious: neuroimaging is one of the domains that forces the language to treat uncertainty as part of the executable system rather than as an explanatory footnote after the fact.