How It Works
Qlro uses WCPP (Workload-Conditioned Physical Projection) — a scoring framework that evaluates quantum devices based on physics, not marketing.
Every score is computed from real benchmark data published by Metriq (Unitary Foundation, CC BY 4.0), anchored to a specific git commit for full reproducibility.
The 3-Stage Pipeline
Physical Translation
Raw benchmark results become dimensionless values in [0, 1]
Each benchmark (e.g., gate error rate, coherence measurement) is transformed into a physically meaningful quantity bounded by a theoretical maximum, not by another device's performance. This is why scores don't change when you add or remove a device from the comparison.
Example: A 2-qubit gate error rate of 0.007 becomes F = 1 - 0.007 = 0.993, bounded by the theoretical perfect fidelity of 1.0
Axis Aggregation
Transformed values are grouped into four capability axes
Multiple benchmarks feed into each axis via geometric mean. If a device has no data for an axis, a conservative population prior is used — and the uncertainty is inflated to honestly flag the gap.
How well qubits can interact. High Γ means less routing overhead for complex circuits.
How long qubits stay usable. Critical for deep circuits that need sustained quantum states.
How accurate each operation is. Per-gate errors compound — high F means less noise accumulation.
How fast the device executes. Matters when running many shots or iterative algorithms.
Workload Composition
Axes are combined with workload-specific weights
The final fit score is a weighted geometric mean across the four axes. The weights depend on your workload — a chemistry simulation weights fidelity and coherence heavily, while an optimization problem cares more about connectivity and throughput.
fit(device) = Γw₁ × Φw₂ × Fw₃ × Tw₄
where w₁ + w₂ + w₃ + w₄ = 1 and weights are set by your workload spec
The geometric mean ensures that a device with any single weak axisgets pulled down — you can't compensate for terrible coherence with great fidelity if your workload needs both. This reflects the physical reality of quantum computing: capabilities are complements, not substitutes.
Provable Properties
WCPP isn't a heuristic — it has four mathematically provable properties:
Baseline Invariance
Adding or removing any device from the comparison leaves all other scores unchanged. Your recommendation doesn't shift because a new device entered the market.
Workload Discrimination
Different workloads produce different rankings. A chemistry workload and an optimization workload rank the same devices differently — because they have different requirements.
Monotonicity
If a device improves on any benchmark, its score can only go up (never down). Better hardware always means a better score.
Bounded Output
Every score falls in [0, 1] with a clear interpretation: 0 means completely unsuitable, 1 means theoretically perfect on all axes the workload cares about.
What the Fit Score Does NOT Include
Transparency means telling you what's excluded:
Cost: Cloud pricing, queue times, and access tiers are not part of the score.
Availability: Whether you can actually access the device right now.
Software ecosystem: SDK maturity, documentation quality, transpiler support.
Future roadmap: Planned upgrades, qubit count expansions, error correction timelines.
The fit score measures physical capability match only. Practical deployment decisions should combine the fit score with these external factors.