I found that a test command can pass without actually running a declared test, so I built a stricter evidence model
I’m building an open-source tool called ProofDiff that analyzes a code change and tries to show what verification evidence actually exists.
While testing it, I found an assumption I had made was wrong:
node --test helper.js can exit successfully even when the file doesn’t contain a declared test.
My original implementation could therefore treat a successful targeted command as stronger evidence than it really was.
I changed the model so a related test only strengthens the result when ProofDiff can establish:
static relationship → qualified test target → exact target executed → runner observes at least one real non-skipped test → pass
A successful process exit alone is no longer enough.
The project is still early and I’m currently improving static dependency resolution for TypeScript path aliases and package exports.
I’d especially appreciate feedback on the evidence model or cases where this approach might still overstate what was tested.