I built something that decides when it should stop and pass a case to a human instead of guessing. I can measure whether it is accurate. I cannot measure whether its decisions look sensible to someone who has actually seen patients.
No right answers, and nothing here is a test of you. Each case shows what was written on the record, what my thing decided, and the questions it says to ask next. Then two taps.
Two of the thirty are cases where it got it wrong, and I am not saying which.
Finding one is the most useful thing that could happen to this.
It is a student project — not a medical device, not approved for anything, and nobody will be
treated on the strength of it.
So I describe you accurately — “one final-year medical student reviewed 30 outputs”, never “clinically validated”. Credited however you like, including not at all.
Written up exactly as you said it, including anything unflattering. Credited however you want, or not at all. Described precisely — one reviewer is one reviewer.
Four of the thirty cases are not fully blind: the record itself carries a history question naming the condition. That is real clinical information rather than a leak, and the write-up will say so.