How it works

A nameplate photo (or typed fields), condition photos, answers to a few symptom questions, and any bench/field test readings. Every field is optional — the read sharpens as you add evidence.

Perception reads each input with its own confidence and deterministic physics checks catch impossible values; then one reasoner fuses the channels — agreement raises confidence, conflicts are surfaced, and a diagnosis is never more certain than the evidence under it. A hard gate enforces honesty and an adversarial pass argues the alternative.

A ranked differential, an honest 'decidable now vs needs a bench test' call, the single best next step, and the most useful inputs to add. Your expert confirms or corrects — and that correction teaches the system. (v1 does not quote.)