Rebound

Navigation

Explore

Choose a platform module or vertical, then jump straight to the corresponding page.

Platform / Module 03

Horizon-Aware
Selector

The decision engine. Every inference request carries a target horizon k and a risk function. The selector asks: which model in the Rashomon set maximizes expected decision utility— not raw accuracy — at that horizon? And when k > k*, it refuses.

selections / new
Latency: 40ms
Selection request
Pool
wind_dispatch
Horizon k (steps)
12 · 12h
Current k*
14 ✓ ok
Risk function
balancing_penalty · €/MWh
Regime at request time
Transitionalλ = 0.087
Result
Allowed
reservoir-net
selected · horizon k = 12
reservoir-netutility 0.94 · rmse 0.045
cnn-lstmutility 0.81 · rmse 0.044
transformer-v3utility 0.71 · rmse 0.043
llm-finetuned-butility 0.68 · rmse 0.046
gbdt-baselineutility 0.52 · rmse 0.047

Rationale: Reservoir computing dominates short-horizon utility in the current Transitional regime. The default RMSE-optimal model (transformer-v3) ranks 3rd once balancing penalty is the objective.

audit_id = ax_8f3c2b91fe04 · signature verified · latency 28ms

Blocked · k > k*
A parallel request at k = 22 was refused at 12:13:48 UTC.

No model in the pool meets the utility floor beyond the predictability horizon. The call returned a structured refusal, not a hallucinated answer.

audit_id = ax_a11b00dd

Decision utility, not RMSE

Accuracy is a proxy. We maximize the domain-specific cost function — €/MWh, basis points, defect-weighted edit distance, expected kill-chain error.

Context-aware routing in < 40ms

Each request carries k, regime, and risk. The selector pulls per-model reliability at that horizon from cache and decides in under 40ms at p99.

Refusal as a first-class response

When k > k* or the utility floor fails, we return a structured REFUSED envelope with reason codes. Downstream systems stop; they do not plow ahead.

Route every inference through a trust gate.

Python SDK, REST, and gRPC ready on day one.