Vancomycin Uncertainty
Full uncertainty propagation and compile-time refusal for vancomycin dosing.
Vancomycin Uncertainty
This is the strongest public case study on the docs site because it shows why Sounio exists: confidence, evidence, and provenance are not comments around the program; they are part of whether the program is allowed to proceed. The current repo still proves that point with real fixtures and screenshots.
Core proof points
tests/run-pass/vancomycin_propagation.sioshows the lighter-weight current propagation path.tests/run-pass/med/vancomycin_full_propagation.siois the fuller clinical kernel-style example in the med lane.tests/compile-fail/vancomycin_low_conf.sioandtests/compile-fail/med/vancomycin_low_conf_refusal.sioprove that low-confidence states can be rejected before runtime.tests/compile-fail/med/vancomycin_weak_evidence_refusal.sioshows that evidence-tier constraints also matter.
Validation commands
export SOUC_BIN="$(pwd)/artifacts/omega/souc-bin/souc-linux-x86_64-jit"
export SOUNIO_STDLIB_PATH="$(pwd)/stdlib"
"$SOUC_BIN" check tests/run-pass/vancomycin_propagation.sio
"$SOUC_BIN" check tests/run-pass/med/vancomycin_full_propagation.sio
"$SOUC_BIN" check tests/compile-fail/vancomycin_low_conf.sio
"$SOUC_BIN" check tests/compile-fail/med/vancomycin_low_conf_refusal.sio
"$SOUC_BIN" check tests/compile-fail/med/vancomycin_weak_evidence_refusal.sio
What this page demonstrates
- Confidence thresholds can become compile-time contracts.
- The refusal path is part of the verified public story, not just a research aspiration.
- The med fixture family ties the broader epistemic model to a domain where wrong confidence handling is not a cosmetic bug.
Visual evidence

How to discuss it responsibly
- Describe the compile-time refusal path as verified via
check. - Be explicit when a larger runtime or self-host execution path still has limitations.
- Use the case study to explain why Sounio couples types, provenance, and confidence instead of treating them as separate subsystems.