The QNACA runtime, running
A dispatcher for CPUs, GPUs and the quantum processor you have just taken apart. Each task is compiled into several plans; the ones that cannot work are struck out with a stated reason; the survivor runs, and its answer is released only if it passes the original verifier.
Changes apply to the next task. Try Today with the error at 10⁻⁵ and a 5 s queue to find where the quantum branch first survives.
- Computed: every Max-Cut instance is solved exactly, by local search and by annealing in your browser, and QAOA is simulated on a statevector when it runs.
- Modelled: wall-clock time, from the seven-term sum and the parameters above; scheduling and spin-dynamics tasks.
Waiting for the first task…
Geometric mean per task, with the total in brackets. The oracle picks the best plan with hindsight and cannot be deployed; the gap between rule and oracle is all a router can ever win.
Nothing rejected yet.
Problem features, candidate estimates, current backend state and user policy.
A ranked feasible plan, an uncertainty estimate and a fallback.
Filter impossible plans first, then predict. It must be able to abstain and choose a dependable classical strategy.
- A valid specification exists before any solver is invoked.
- Every transformation records its assumptions and a decoder.
- Every accepted result passes the original task verifier.
- The selection policy cannot relax a quality target to improve its reward.
- Provider capabilities and budgets are checked before submission.
- Project QNACA Research and Development Blueprint (12 November 2026)· checked
- Xu, Hutter, Hoos & Leyton-Brown, SATzilla: portfolio-based algorithm selection, JAIR 32 (2008)· cited in blueprint
- Quantum Portfolio Optimization: An Extensive Benchmark (2025)· checked
- Comparing classical and quantum QUBO solvers (October 2026)· checked
- Moussa, Calandra & Dunjko, To quantum or not to quantum (2020)· checked
- Lindauer, Hoos & Hutter, From sequential algorithm selection to parallel portfolio selection (2015)· checked