triage-quantum-use-case

Skill: triage-quantum-use-case

Invoked by: quantum-solutions-architect (primary). Also consulted by quantum-algorithm-engineer when a build reveals the problem can't fit the depth/coherence budget and needs a re-triage.

When to invoke: "should we use a quantum computer for this?"; "gate-model or annealing?"; "which qubit modality?"; "is this NISQ-doable or does it need fault tolerance?"; "how many qubits would this take?"; "Qiskit / Cirq / PennyLane / Braket — and whose hardware?"; any "is this a quantum problem, and how would we approach it?" question.

Output: a go/no-go triage verdict (defaulting to "classical wins today") + IF it survives: the paradigm + qubit modality + NISQ-vs-FT placement + order-of-magnitude resource estimate + SDK/provider + the flip conditions — every volatile fact dated and marked verify-at-use.

Procedure

1. Restate the problem in the tree's terms. Capture: the workload (combinatorial optimization / quantum chemistry / linear systems / factoring-cryptanalysis / search / machine learning / sampling), the problem size (variables, qubits it would need), the data-loading shape (does the algorithm need a large classical dataset loaded into a quantum state?), and the best classical method it would have to beat.
2. Run the triage gate FIRST — this is the whole point. Three killer questions, all must be "yes" to proceed:
   • (a) Proven advantage? Is there a known quantum algorithm with a proven asymptotic advantage for this problem (not a hope, a theorem)? Chemistry/simulation and factoring have one; generic business optimization mostly does not.
   • (b) Size in the advantage regime? Does the advantage appear at problem sizes reachable on real hardware, or only asymptotically at qubit counts that don't exist? Many advantages are "someday," not "now."
   • (c) State-prep affordable? Does loading the classical data into a quantum state (state preparation) cost less than the speedup buys? An exponential algorithm behind an exponential data-load has no advantage.
   • If any answer is no → verdict: "classical wins today — revisit as hardware matures." Say it plainly and name the classical baseline. Most enterprise asks end here, and that is the correct, valuable outcome.
3. Only if it survives, choose the paradigm. Traverse ../../knowledge/quantum-computing-decision-tree.md: QUBO/Ising optimization → quantum annealing or gate-model QAOA (and still check classical solvers first); universal algorithms (chemistry, factoring, HHL, search) → gate/circuit model; specific photonic/cluster-state schemes → measurement-based.
4. Choose the qubit modality as a trade-off. Superconducting (fast gates, short coherence), trapped-ion (high fidelity, all-to-all, slower), neutral-atom (reconfigurable, scaling), photonic (room-temp, networking, probabilistic gates). Frame fidelity-vs-connectivity-vs-speed-vs-scale for this workload; mark specific qubit counts/fidelities verify-at-use.
5. Place it on the NISQ-vs-fault-tolerant timeline honestly. Needs error correction (Shor / HHL / QPE) → fault-tolerant → not practical yet; shallow variational (VQE / QAOA / quantum kernels) → NISQ-doable with error mitigation only. Name the physical-vs-logical qubit distinction and the surface-code overhead if FT.
6. Estimate resources at order-of-magnitude. Physical (and logical, if FT) qubit count, circuit depth vs coherence budget, shots for the target statistical error. Often the estimate is the verdict ("needs millions of physical qubits" = no-go today).
7. Select the SDK/provider and demand simulators first. Qiskit/Qiskit Runtime (IBM), Cirq (Google), PennyLane (hybrid/differentiable, hardware-agnostic), Braket (multi-provider). Prove on a simulator before any QPU spend. Name the queue/cost reality.
8. State the flip conditions — usually a hardware milestone (e.g., "if a device with N logical fault-tolerant qubits ships, this no-go becomes a go"). Date every volatile claim.

Worked example

User: "We want to optimize our delivery-truck routing with a quantum computer. QAOA on a real QPU — which one?"

• Workload: combinatorial optimization (vehicle routing / a QUBO). Best classical baseline: mature solvers — Gurobi/OR-Tools, LKH, simulated/parallel-tempering annealing — that solve industrial routing at scale today.
• Triage gate: (a) proven advantage? No — QAOA has no proven asymptotic advantage over the best classical heuristics for generic routing. (b) size in regime? The instances that matter need far more clean qubits than NISQ offers. (c) state-prep? The QUBO encoding is large and the results are noise-dominated at useful sizes.
• Verdict: CLASSICAL WINS TODAY. Recommend a classical solver now; revisit QAOA only as a research track, not a production plan. This is the honest, valuable answer — the team saves the QPU budget.
• Flip condition: if fault-tolerant hardware with thousands of logical qubits and a demonstrated routing advantage over the best classical solver appears, re-triage. (Every hardware claim here is dated and verify-at-use.)

Guardrails

• Run the triage gate before anything else — never brand-match a quantum solution to a request that classical wins. "Classical wins today" is the default, not a cop-out.
• Compare against the BEST classical method, not a strawman. An advantage over a weak baseline is not an advantage.
• Always check the state-preparation cost — an exponential speedup behind an exponential data-load is no speedup.
• NISQ ≠ fault-tolerant. Don't place a Shor/HHL/QPE problem as if today's hardware could run it; say "not practical yet."
• Modalities are trade-offs, not winners — name what each buys and costs for the workload.
• Post-quantum cryptography migration is NOT this skill — that's the defensive switch to quantum-resistant algorithms → route to security-engineering / cybersecurity-grc.
• Every qubit count, fidelity, coherence time, and SDK/provider fact carries a retrieval date + [verify-at-use] — this landscape moves monthly. See ../../knowledge/quantum-computing-patterns-2026.md.