Arm A — the model scores 15 names and we read out the top two

Arm A — the model scores 15 names and we read out the top two A workflow diagram generated by Archify. 01 / Farm Operator 02 / Working it out 03 / Notes and the exit Listen What to ask What to say Hear · Farm Operator writes in · Farm Operator › Listen · one of four paths Hear Farm Operator writes in one of four paths Ask it · every blank in one go · Farm Operator › What to ask · seen: 7 questions Ask it every blank in one go seen: 7 questions Say it · reads out the top two · Farm Operator › What to say Say it reads out the top two Build the list · the same 15 names · Working it out › Listen · 14 diseases + None Build the list the same 15 names 14 diseases + None Pick a question · whatever is still blank · Working it out › What to ask · a fixed checklist Pick a question whatever is still blank a fixed checklist Stop or go on? · when nothing is blank · Working it out › What to say Stop or go on? when nothing is blank Allowed to say · 14 paid model calls · Working it out › What to say · always the top two Allowed to say 14 paid model calls always the top two Send to the vet · None must beat all 14 diseases · Notes and the exit › What to say · so it is not really a door Send to the vet None must beat all 14 diseases so it is not really a door Write it down · only what each call cost · Notes and the exit › What to ask · no reason kept for a skip Write it down only what each call cost no reason kept for a skip ranks 1 and 2 still blank? ask again refuse by score cost Legend Arm A's own design A refusal that must win a contest The box to look at

What makes Arm A different

  • • This is the starting design. Nothing came before it.
  • • A model gives each of 15 names a score.
  • • We read out the top two, whatever the scores say.

Why that is risky

  • • The words Very High and High come from cutoffs with no source
  • • It names two diseases every time, even when it is unsure
  • • To say I do not know, None has to beat all 14 diseases

Before you quote this

  • • The Arm A we measured was a simpler stand-in, not this one
  • • Research only. Nothing is deployed.
  • • No percentage reaches the Farm Operator.