The same household, two ways
On the left, the process as it exists: one application per program, each asking for most of the same facts again. On the right, one description and every rule checked together. Both sides ask about exactly the same facts: the questions on the left are generated from the same rule library, so the repetition is real rather than staged.
The way it works today
21 applications · 54 questions
Three programmes, three applications. 54 questions in total, asked one form at a time, the way they are today.
The way BennyFit works
One description, every rule checked together, and a question only where it changes the answer.
Typed readout
the engine · one batched pass, no text generated
Not run yet.
Generation lanes
not run yet
Two lanes run the identical household and identical criteria through a model asked to write JSON. One holds the model fixed and changes only the readout method; the other changes the model as well, so “a better model would not do that” can be answered rather than argued.
Screening estimates from published federal and state rules, not eligibility determinations. BennyFit keeps nothing typed here, but it is sent to our AI provider to be read, so please describe a made-up household. Use the full demo · How accuracy is measured