Geometry
I checked Lovibond angle, depth ratio, and related math on known coordinates.
Research
Engineering validation
I record what the software already does, and what my automated tests actually checked. A passing test is not diagnostic accuracy.
Illustration of engineering checks. Not a clinical accuracy result.
What I have tested
Seventeen automated tests across these suites, dated 2026-10-19. Summary: test summary. Full log: test log.
Quality
I reject a finger that is too small, blur, motion, and extreme exposure. An aligned fixture passes.
Fusion and the one-image rule
I reject a one-image patient conclusion. I also exercise the multi-finger path.
Stub withhold
I check that labels stay consistent with keeping stub results research-only. A stub may not look clinical.
What is not established
Passing engineering tests does not prove clinical performance.
Clinical accuracy
I have not established diagnostic utility, population performance, or calibrated uncertainty. Classification bands in code are provisional.
Fairness
I have not established fairness across skin tones. I do not claim demographic fairness from these tests.
Landmark AI
Trained landmark models are not the default path. A number from placeholders is not proof I found the anatomy.
Cameras and lighting
Do lighting, pigmentation, and camera quality change landmark error, and therefore the angle or depth-ratio numbers? I have not proven the system across devices.