Workstream C · Research dossier

Quantum control
and PERSIST-Q.

PERSIST-Q investigates whether a prospective estimator can identify useful intervention opportunities and improve quantum-control performance under realistic resource and latency constraints.

Proposed protocol

The decision

What each intervention involves

Information before intervention

The observations, sensor readouts and records available at the decision deadline.

Available action

The correction the controller may apply, for example phase feedforward, or no action.

Outcome measure

The later readout or fidelity measure that shows whether the action helped.

Cost of intervention

Measurement, reset, actuation time and any added error.

Competing controller

The strongest implementable conventional controller acting on the same information.

No public specification of a PERSIST-Q estimator is available on this website, so no estimator equation is shown. An estimator will be displayed only with its actual definition, inputs and source version.

Evaluation design

Comparators and matching

Compare

The strongest implementable conventional controller.
Relevant simpler baselines.
The proposed PERSIST-Q controller.

Comparators are chosen for the actual dataset and control problem. Candidates may include OU or AR models, Kalman filtering, hidden Markov models and suitable learned predictors.

Match

Available observations and sensors
Measurement and reset resources
Training information and tuning budget
Intervention opportunities
Computational and communication latency
Decision deadlines
Evaluation data

Report end-to-end latency, including measurement, transfer, computation and actuation, and count deadline misses rather than reporting computation time alone.
Use a held-out evaluation. Where several devices or backends exist, test transfer to an unseen backend or device.
Claims of intervention benefit need paired or randomised evidence, or an explicitly identified simulator with its assumptions. An observational association is not a demonstrated causal improvement.

Three different claims

Prediction, benefit and improvement

Predicting an outcome.
Predicting whether an intervention will help.
Delivering a measured net control improvement.

Each needs its own evidence. An engineering advantage would not, by itself, experimentally establish PDT’s full foundational ontology.

Dossier summary

Protocol, evidence and criteria

Proposed evaluation

A reproducible comparison against the strongest implementable alternatives using suitable data and a fixed protocol.

Inputs and assumptions

A control problem, dataset or device, a specified estimator with its source version, matched comparators, resource and latency budgets.

Proposed measurements

Net task performance, end-to-end latency, deadline misses and intervention rate on held-out data.

Evidence currently available

No benchmark report is available. No control advantage is claimed. The result still required is a benchmark report under a fixed protocol.

Comparison or rejection criterion

A net improvement over the strongest matched controller on held-out data, including end-to-end latency and deadline misses, with paired or randomised evidence for intervention benefit.

Reproducibility

No dataset, device record or code is published. A benchmark report must give data provenance, splits, exclusions, metric definitions, sample counts and uncertainty.

Current foundational references

R04 R06