Spend each simulation on the answer, not setup.

Use one vector response to produce the next complete update.

Abstract gradient field
10 / 10

fixed systems

six charged responses

9 / 10

changing systems

six charged responses

1

response per update

verification read separate

Connect measure(x) to the response vector your simulator already returns.

Call once, pass the response to step(), then evaluate the returned configuration.

Open the quickstart

Does your simulator expose the response component by component, and is each call expensive enough that evaluation count dominates?

Check compatibility

Where it applies

Response matchingCoupled model calibrationDigital-twin trackingHardware-in-the-loop controlNonlinear least squaresInverse response problems

What this evidence shows. And what it does not.

Controlled fixed and changing vector-response suite.