Unit of Assurance
Unit of Assurance
Machine-verifiable credibility evidence for regulated computational simulation. One signed JSON-LD package per Context of Use.
$ pip install uofa $ uofa demo ══ UofA Demo — Centrifugal Blood Pump CFD ══ C1 + C2 + C3 pipeline · pre-computed artifact ✓ C1 Integrity verified hash + ed25519 signature valid ✓ C2 SHACL validation Minimal profile conforms ⚡ C3 Quality gates 7 weakeners detected └── W-AL-01 ×2 · W-AR-05 ×2 · W-EP-02 ×2 · W-ON-02
What it is
A decision packaging construct, not another process framework.
Decisions as artifacts
Sign credibility decisions with ed25519. Hash with SHA-256. Make the decision portable, tool-independent, and tamper-evident.
Read more →Computable profiles
SHACL profiles enforce what a UofA must contain. Minimal for live capture. Complete for regulatory submission.
Read more →Defeater detection
Forward-chaining rules catch orphan claims, missing UQ, and compound credibility risks no SPARQL query can find.
Read more →Pluggable domain packs
ASME V&V 40 for FDA medical device. NASA-STD-7009B for aerospace. DO-178C planned. Bring your own.
Read more →From evidence to signed package
From an evidence folder to a signed package.
You don't start in JSON-LD. You start with the PDFs, reports, and configs you already have. Extract proposes a first pass, you review it, and one command signs and validates the result.
Drop your evidence
Source PDFs, validation reports, solver configs, and acceptance criteria into one folder. No layout required.
uofa extract
A model reads the folder and pre-fills the credibility-factor workbook. One sheet per factor. No RDF knowledge needed.
Open the workbook
Confidence coloring shows where the model was unsure. You correct what it got wrong. This is the checkpoint, not a formality.
uofa import
Reviewed workbook to signed, validated JSON-LD in one command. SHACL completeness and integrity checks included. About 30 seconds.
uofa rules
Forward-chaining rules flag weakeners and compound credibility risks across the signed package.
$ uofa extract ./evidence --pack vv40 -o assessment.xlsx → 13 factors proposed · review confidence flags before import # ... you review and correct assessment.xlsx ... $ uofa import assessment.xlsx --pack vv40 --sign --key research.key --check ✓ signed · SHACL Complete profile conforms $ uofa rules assessment.jsonld --pack vv40 ⚡ 3 weakeners detected
Extract proposes. You verify. Nothing is authoritative until you sign it. Run extract locally with Ollama and your evidence never leaves your environment, or point it at a hosted model if you prefer.
Live demo
Morrison blood pump,
re-expressed in 30 seconds.
Morrison et al. (2019) is the FDA-co-authored worked example for ASME V&V 40. The prose paper deems the model credible for COU 1 and inadequate for COU 2. Re-expressed as UofA evidence packages, the same decisions become signed, hashed, and machine-verifiable.
Same model. Same data. Different model risk. The rule engine surfaces the risk-driven divergence automatically.
$ uofa rules packs/vv40/examples/morrison/cou2/uofa-morrison-cou2.jsonld ══════════════════════════════════════════════════════ UofA Weakener Detection Report Input: uofa-morrison-cou2.jsonld ══════════════════════════════════════════════════════ SUMMARY: 18 weakener(s) detected ───────────────────────────────────────── Critical: 9 High: 7 Medium: 2 ⚡ W-PROV-01 [Critical] 7 hits ⚡ COMPOUND-01 [Critical] 2 hits → cou2 ×2 ⚠ W-EP-04 [High] 6 hits ⚠ W-ON-02 [High] 1 hit → cou2-vad ⚠ W-AL-02 [Medium] 1 hit → cou2 ⚠ W-CON-04 [Medium] 1 hit → cou2 ───────────────────────────────────────── ⚡ 2 compound inferences — chained reasoning, not detectable by standalone SPARQL.