Overview

The drop-in maintenance layer and stability contract for component-observable continuous systems.

TrueLoop sits between a digital host and a photonic, wave, quantum, RF, sensing, or in-memory analog substrate. The host keeps orchestration and semantics. The physical layer keeps the dense transform. TrueLoop keeps the delegated physical state inside a declared operating tolerance.

The complete product has four parts: exactly budgeted preflight, a five-state stability contract, evidence-scoped controller-family guidance, and an Ed25519-signed run record. The controller is one member inside that layer, not the whole product.

Each ordinary cycle consumes one current full response vector and returns one complete writable configuration. The active default performs linear host work and carries bounded auxiliary state per writable control. Flat scaling refers to sequential acquisition depth, not zero traffic or zero arithmetic.

Local preflight fit requires a component-resolved response, a measured feasible target, same-cycle timing, and declared configuration bounds. The registry separately reports evidence-backed fit or a monitored boundary pilot. Unknown evidence is never promoted to evidence or converted into a universal decline.

Per-channel closed-loop drift correction interleaved with a workload is established: Vepsäläinen et al. (Nature Communications, 2022) stabilised a single transmon this way and noted it should extend to many qubits at no added cost; spectator-qubit monitoring (Majumder et al., 2020) and real-time Hamiltonian estimation (Shulman et al., 2014) are the same idea on other platforms. TrueLoop's contribution is not the loop. It is the contract around it: a substrate-agnostic maintenance layer with a stated host-side cost model, formal preconditions for when it holds, a certification that records what it has and has not demonstrated, and the demonstration that this suffices at scale (96 channels of native drift for nine hours) and transfers across analog, photonic, thermal, and trapped-ion plants without retuning.

The layer sits underneath the error-mitigation stack rather than beside it. Zero-noise extrapolation scales noise by gate folding G → G†G G; a coherent over-rotation in G† carries the opposite sign and cancels in the fold. Measured: with coherent error only, ZNE returns identical values at fold factors 1, 3, and 5 and extrapolates to a confidently wrong number. Stochastic error folds correctly. So coherent drift has to be removed at the source, and the mitigation stack handles what remains.

layer = StabilityContract(key, n=len(x0), mode="regulation", target=target)
x = layer.start(x0)
for _ in range(rounds):
    x = layer.step(measure(x), target=target)
proof = layer.end()