Uncertainty

How Sounio represents and (eventually) propagates measurement uncertainty.

Uncertainty

Scientific code is full of measurements, and measurements are never exact. Sounio treats uncertainty as first-class data instead of something you track in comments or spreadsheets.

What Works Today (Compiler Reality)

The compiler enforces the “no silent unwrap” rule for Knowledge<T> (see Knowledge<T>). That gives you a reliable boundary where epistemic information is intentionally discarded.

Beyond that, full automatic uncertainty propagation depends on how much of the epistemic stdlib you use, and which compiler features/backends are enabled.

In the Spec / Intended Semantics

The intended model follows standard metrology practice (GUM-style propagation for independent inputs):

  • For c = a ± b:
    • u(c) = sqrt(u(a)^2 + u(b)^2)
  • For c = a * b or c = a / b (relative form):
    • u_rel(c) = sqrt(u_rel(a)^2 + u_rel(b)^2), where u_rel(x) = u(x) / x

At the language level, the goal is that operations on Knowledge<T>:

  • compute the new point estimate
  • propagate uncertainty
  • update confidence (when applicable)
  • append provenance

Practical Guidance

  • Keep values as Knowledge<T> as long as you can.
  • When you must cross into “plain data”, use unwrap(reason) and record why.
  • For domain code (PK/PD, neuroimaging, etc.), prefer the stdlib modules that already model uncertainty and provenance explicitly.

Next