The instrument, not the pitch
See the measurement actually run.
The browser-only assessment demo selects from 16 illustrative items using a 2PL information function and reports an unnormed estimate with an uncertainty band. It does not produce a validated CAI score.
Working demo · runs offline
A live scoring demonstration.
Eight reasoning items, chosen adaptively. Each answer updates a Bayesian ability estimate, and the next item is selected for information at the current estimate. The parameters are illustrative and the output is not a validated CAI score.
Adaptive selection
Each item is picked for maximum information at your current estimate.
Honest error bars
Every score carries a confidence band. It never claims certainty.
Trajectory-first
One point is nearly meaningless. The unit that matters is change over time.
No verdicts
Illustrative calibration, not a diagnosis. The math is real; the item norms are a placeholder.
The rigor behind it
Validation architecture
The measurement model, how change is told apart from noise, equating, longitudinal invariance, fairness testing, and where the neural anchor stops.
Read the method→The construct
Cognitive debt and reasoning autonomy
The flagship paper: what reasoning autonomy is, why it is now measurable, and what the instrument can and cannot claim.
Read the paper→