When does a quantum method deserve another experiment?
A promising result leaves a practical question: should we keep going? Four checkpoints help us decide what to test before investing more time.
An encouraging experiment creates momentum. The next run is going to be larger, the question more ambitious, the hardware time more expensive. It is easy to keep going because stopping feels like losing what we have learned. I prefer to pause while the result is still fresh and ask what the next experiment would let us decide.
In the NODERIQ research plan, I use four checkpoints, labelled G0 to G3. They follow the work from a clear question to a possible practical contribution. A negative answer is useful too: it can show where to change the question or stop.
G0: What are we trying to get done?
What is the input, what answer do we need and how long can we wait? If the task cannot tolerate remote hardware access or cannot leave local control, that matters before choosing an algorithm. A precise question can save a great deal of experimentation in the wrong direction.
This is also the place to agree how success will be measured. A task assignment problem, for example, still needs to respect the original constraints. A faster answer to an easier version is not the answer we set out to find.
G1: What are we comparing it with?
Before hardware, the method should give valid simulated results against a capable classical reference. For QANTIS belief updates, that reference is exact Bayesian inference. The comparison needs the method an engineer would genuinely consider using, not a weak option chosen to make the alternative look good.
If it cannot do that yet, the simulation has still helped. It gives us a cheaper place to understand the problem before adding queue time, calibration and hardware noise.
G2: What changes on the hardware?
The result has to preserve the information and uncertainty the task needs. A method might look better after quietly dropping a difficult constraint. That is why I want to inspect what was represented at the input and what remains at the output, as well as the headline score.
The QANTIS hardware campaigns were built to investigate this stage for the declared workloads. Their results help describe particular conditions; they do not settle every future use of the method.
G3: Does the whole job benefit?
Now count accuracy, waiting, reliability, cost and the effort of running a second path. A more accurate component may still be the wrong choice for a job with a short deadline. The useful outcome is a decision about the whole workflow, not only the circuit.
For the belief-update work, G0 and G1 have been addressed, and the hardware studies investigate G2. G3 remains open. NODERIQ therefore starts with a useful classical AI core, while quantum assistance remains a conditional research path. The checkpoints let us explain that choice in terms another person can examine.
- Define and simulate: G0, G1
- Inspect the hardware result: G2
- Evaluate the whole job: G3