Quantum architecture program — what we actually tested

This page explains, in plain language, a focused IBM Heron program (June 2026) to show that Quantum Polycontextural Computing (QPC) is a real gate-level architecture — not “run three cloud jobs and merge in Python.” The goal is demonstration and falsification, not claiming to beat published QAOA scores or win vendor comparisons.

What QPC means here (one sentence)

Multiple objective “contextures” execute in parallel inside one quantum circuit, linked by transjunction gates while alternatives stay open — then you measure once on the gate QPU.

That is different from sending carbon, biodiversity, and social impact as three separate IBM jobs and combining results classically.

Why an external AI review mattered

An independent Claude review initially treated stacked Hamiltonian layers as “QPC.” We corrected that: sequential layers are not parallel contextures. After seeing true parallel circuits and IBM job IDs, the reviewer acknowledged the gap and proposed a separability falsification test — measure correlation that independent jobs cannot reproduce.

We implemented that test, ran it on ibm_fez, and document every outcome honestly (pass, partial, and closed cases).

Three workstreams — what each one proves

WorkstreamQuestionStatusWhat it demonstrates
Separability (ICC) Does one coupled job produce joint structure that three independent jobs cannot? Demonstrated QPC is not classical multi-job orchestration. Toy 3/3 seeds; Cerrado n=12; Heron n=28 per-boundary pass.
Cerrado portfolio Can real Goiás carbon-credit data run as one polycontextural workflow? Demonstrated Same published metric, open data, auditable IBM jobs — one submission vs three. Scores vary with depth; architecture stands.
Antenna mAPP Can strict one-hot placement work with transjunctions on assignment lines? Closed True parallel K=6 proved (~23% argmax feasible); strict acceptance 0%. Wrong problem class without protected registers — documented internally.
E8 lattice test Can eight parallel “certificate checks” (math-style) run as one coupled Heron job? Demonstrated 152Q one job; E8 as calibration reference only — not a theorem claim. Plain-language report on site.

Results in plain language

1. Separability — “Is it really one quantum thing?”

We run two arms on the same IBM chip:

We compute inter-context correlation (ICC) — roughly: “do decisions on one objective line up with the next more than random independent runs would?”

TestICC gap (coupled − separable)Verdict
Toy instance (seeds 7, 11, 42)0.052 – 0.106Strong pass 3/3
Cerrado real data n=120.088Pass
Cerrado Heron n=28 (mean)0.031Below mean threshold
Cerrado Heron n=28 (bio↔social boundary alone)0.098Heron-scale pass

Value: This is the architecture signature. It does not mean we “won” an optimization contest — it means the hardware saw coupled quantum structure.

2. Cerrado — “Can a real business task use one submission?”

Brazilian Cerrado: pick towns that jointly score carbon, biodiversity, and social impact (Ribeiro 2026, open GitHub data). Same portfolio formula as the paper.

Value: Reproducible gate-native multi-objective execution with job IDs on file. Honest about NISQ depth limits.

3. Antenna — “What did we learn by pushing hard constraints?”

Leonardo multi-frequency antenna placement needs exact one-hot site assignment. True parallel QPC (K=6) ran on IBM; strict feasible one-hot acceptance was 0% on the case brief.

Value: Parallel contextures work; hard local constraints on assignment registers need different encoding. Case closed — not hidden, not marketed as solved.

4. E8 lattice test — “Can multi-check math-style search run on Heron?”

In 8D, the densest sphere packing (E8) was proved by Viazovska using several coordinated checks — not brute search. QPC runs that computational pattern on IBM: eight parallel certificate contextures in one 152-qubit job. E8 is only a known answer key for calibration; we do not claim a new theorem.

Value: Shows QPC’s multi-context architecture on a structured hard problem — same idea as Cerrado’s three objectives, different domain. Plain-language explainer →

How computing works (shared pipeline)

1. Load published or synthetic instance (Cerrado data, toy weights, antenna graph) 2. Compile K parallel context blocks into one gate-native workflow (proprietary schedule) 3. Run coupled (one IBM job) vs separable control (matched multi-job baseline) 4. Primary metric: inter-context correlation (ICC) gap on hardware counts 5. Secondary metrics: portfolio / placement scores with standard decoders (separate claims)

Compilation details, bridge wiring, and readout transforms are proprietary — contact readytogo@quantumpolycontextural.ai under NDA.

What we do not claim

Reports & upload pack

Separability — full results & methodology Cerrado — optimization architecture compare Real-world replication hub Comparable benchmarks index E8 lattice extremal report Hostinger upload list (June 2026)