Pasiflora AITraining Environments
A model gets better by attempting a task thousands of times and being told each time whether it succeeded. Nobody can grade a million attempts. So the grading has to be a program that runs unattended.
An expert task and an environment are one step apart
Our experts already write problems that end in an exact, checkable answer. An environment asks for the same thing one step further along: not the result, but the work that produces it.
The task · what an expert writes today
“Calculate the Year 3 debt service coverage ratio for this leveraged buyout.”
Graded by: did the model say 1.40? One check, right or wrong.
Three parts
The task
A problem written by someone credentialed in the field, precise enough that a machine can decide whether it was done correctly.
The sandbox
An isolated container with the real tools a professional would use, and no network access, so the score keeps measuring capability.
The verifier
A program that returns a score between 0 and 1. No human grader, no second AI acting as judge. It gives the same answer today as it will in a year.
Every field invites its own shortcut
Every task we write ships with deliberately broken answers that have to fail. If one of them scores full marks, the task does not ship.
Mechanical drawing
The shortcutMemorise the coordinates instead of reasoning about the geometry.
What stops itDimensions vary per task, so memorising does not pay.
Financial modelling
The shortcutWork the answer out and type the number into the cell.
What stops itThe cell has to hold a formula that computes the right value from the inputs.
Clinical coding
The shortcutList every plausible code so the correct one is somewhere in the set.
What stops itPrecision is scored, not just coverage, and specific miscodes must not appear.
Want one built in your domain?
Tell us the task and how you would know it was done correctly. That conversation is the whole design.