We did the math
so your regulators don't have to.
Rebound sits on peer-reviewed research at the intersection of chaos theory, ensemble model selection, and decision engineering. The numbers below are proven results, not marketing estimates.
reduction in grid instability penalties — wind power scheduling, 24-hour horizon.
improvement in traffic-flow optimization — adaptive signal timing, 48-hour window.
decision-utility lift across domains when horizon-constrained selection replaces RMSE selection.
Lyapunov exponents reliably predict the forecast horizon at which accuracy diverges from decision utility.
Our papers & preprints
The vocabulary we standardize on.
The set of ML models that perform equally well (within tolerance) on a given validation dataset. Named after the Kurosawa film — multiple equally valid accounts of the same event.
A measure of exponential divergence of nearby trajectories in a dynamical system. Positive λ indicates chaos; larger λ indicates faster divergence.
The lead time beyond which an AI's prediction is statistically indistinguishable from random noise, given the current λ.
The phenomenon where multiple equally-accurate models make systematically different predictions on individual cases or future steps.
A domain-specific measure of the real-world value of an AI recommendation — distinct from statistical accuracy. For trading, risk-adjusted return; for grid ops, avoided balancing penalty.
The subset of the Rashomon set that maintains reliable decision utility up to a specific forecast horizon k.
Work with our research team.
We co-author with design-partner customers on domain-specific results.