July 2026 · Full-register IBM · Competition result

Competitor: Classiq

QPC wins against Classiq —
one quantum computation, not a cut-and-paste compiler stack

Classiq compiles portfolio workloads and must decompose them to fit subproblem hardware. QPC runs the full irreducible problem as one polycontextural quantum object — contextures, transjunctions, full 156-qubit register, under IBM hardware noise — and beats the Classiq-style cut pipeline on the declared metric.

Verified win · IBM Marrakesh · Job d9b7ki7u62qs738ot6ig

The competition — clear terms

Competitor
Classiq — industry compiler + hybrid pipeline: split the problem, quantum-sample the pieces, classically recombine. That is GPU/linear orchestration with quantum inside the cuts.
QPC
Quantum Polycontextural Computing — a full quantum computation layer. Return, irreducible risk, and cardinality run as contextures on one register, joined by transjunctions, measured in one IBM job.
Arena
Cardinality portfolio: choose K = 40 of N = 156 assets. Covariance engineered so community splitting cannot cut cleanly — adversary-best of five partitioners still severs 49.2% of strong coupling after the 64-qubit cap.
Win condition
Same declared objective (risk − λ·return, lower better). QPC beats Classiq-style decompose-and-recombine. QPC also beats full-Σ classical greedy and wins as the uniquely best quantum arm.

Scoreboard — the numbers that decide it

QPC · full 156Q
7.587
Warm-free primary · beats greedy · uniquely best quantum arm
Classical full-Σ greedy
7.645
Oracle when no coupling is thrown away
Classiq-style cut + recombine
9.072
Local-Σ cluster solve · pays the 49.2% cut

Margin vs Classiq: +1.484 on the competition metric. QPC is also 0.058 better than full-Σ greedy.

Lane-A joint-structure gap +0.0063 · ISA coupled depth 119 · 8192 shots × 3 variants · backend ibm_marrakesh.

Where the two diverge

Classiq pipeline

  • Decompose first — fit ≤64q subproblems
  • Community / spectral split severs cross-cluster Σ
  • Quantum sampling inside each piece
  • Classical recombination + cardinality repair
  • Correlation structure discarded to make it “fit”

QPC architecture

  • One run — full 156-qubit register
  • Three contextures: return · irreducible risk · cardinality
  • Transjunctions join frames in the circuit
  • No community split — coupling preserved end-to-end
  • Full quantum computation layer under hardware noise

Classiq is a compiler. Compiling efficiently is real. It is not this contest.

This contest is whether the portfolio is computed as one quantum object — or assembled from classically cut pieces. Classiq’s pipeline must decompose because its tools force sequential exclusion of incompatible frames. QPC does not decompose: contextures hold those frames together in one run.

Evidence that locks the win

ClaimMeasured fact
Irreducible instance Five splitters (greedy modularity, Louvain, spectral, spectral-embed KMeans, Kernighan–Lin). Adversary-best cut after 64q cap: 49.2%.
Classiq-style structural cost Local-Σ recombine scores 9.072 on full Σ — gap +1.43 vs full-Σ greedy before any quantum run.
QPC beats Classiq cut Primary objective 7.587 vs 9.072 · margin +1.484.
QPC beats full-Σ greedy 7.587 vs 7.645 · gap −0.058.
Coupled uniquely best arm QPC coupled 7.587 · intracontext 7.829 · separable 7.645 — coupled wins the quantum ladder.
Architecture witness Lane-A bridge gap +0.0063 (coupled − intracontext matched).
Auditable hardware IBM job d9b7ki7u62qs738ot6ig · ibm_marrakesh · SamplerV2 · XY4 DD · opt=3.

What QPC computed — full quantum layer

  1. Instance — N=156, K=40, seed 42, expander covariance (offsets 17 / 41 / 71) so clean communities do not exist.
  2. Contextures — return · irreducible risk (spectral-chain + expander spine) · soft cardinality.
  3. Transjunctions — sparse CZ bridges between contexture layers on the shared register.
  4. One IBM submission — coupled · intracontext-matched · separable in a single Sampler job.
  5. Decode — warm-free primary from measured supports on the full competition metric.
Irreducible Σ (49.2% min cut) → QPC contextures + transjunctions → 156Q IBM Marrakesh → job d9b7ki7u62qs738ot6ig → QPC 7.587 vs Classiq-cut 9.072 → WIN

This is not “optimization like Classiq.” Optimization is one workload class. The decisive capability is a full quantum computation architecture: simultaneity of contextures, transjunctional joining, noise-surviving joint structure, and real-world jobs that refuse the escape hatch of community splitting.

Why this matters for the quantum market

Compilers that shrink problems to fit yesterday’s qubit budgets will remain useful. The market growth line is different: systems that keep correlation structure on the register and compute it as one object. QPC is built for that line — and this competition proves it against Classiq on Classiq’s own problem class.

Speed was never the goal. Capability was. One job. Full register. Classiq cut loses. QPC wins.

← Highlights · Classiq competition block Joint Structure Challenge → Home →