LocalStudy

Every model, evaluated in the open.

One question runs through the Digital Foot Lab: is it defendable? Proven, not promised — deterministic gates, hash-chained receipts, and your photos never leave the box.

FootLab — real foot photos, in the openStudy 002 · model vs model

Gemma 3 27B vs MedGemma 27B

same arch · same prompt · same box · photos stay local

35/35 valid — a tie

Clinician review open

Base Gemma 3 blankets every photo “urgent”; the medical model discriminates — but whether that’s correct is deferred to a blinded podiatrist. The photos never left the box; only the receipts are public. Read the receipts →

01 · Pre-registered

Protocol frozen first

Endpoints and pass/fail thresholds are written and hash-frozen before any results, so nothing is cherry-picked.

02 · Deterministic gates

No model judges a model

Every number is rule-based and recomputable by an outsider. A model never grades a model.

03 · Kept local

Receipts public, data private

Patient photos and observations stay on the owner’s box. Only the aggregate receipts and the tamper-evident ledger are published.