June 2026 · Pre-registered · Public dare

Joint Structure Challenge

Prove on real quantum hardware that one coupled polycontextural job is not the same as K independent cloud jobs merged in Python. Six levels. Auditable IBM job IDs. If you can falsify it, publish your run — that is how the challenge works.

Plain English

What this test is
A public, pre-registered ladder that asks: does one coupled IBM job produce boundary statistics that three separate jobs + Python merge cannot match?
Why we did it
Polycontextural architecture only matters if it is falsifiable on real hardware — not if it is relabelled multi-job orchestration.
What QPC computed
Six levels on ibm_fez: ICC vs separable jobs (L5a), intracontext depth control (L5b), Cerrado merge failure (L4), open vendor dare (L6).
Headline result
L1–L5 pass on raw IBM counts · L5a gaps 0.289 · 0.101 · 0.130 · L5b intracontext 3/3 · auditable job IDs at every level.
What we claim — and do not
Claim: coupled transjunction structure is measurable on NISQ hardware. Do not claim: best portfolio score vs every classical or QAOA baseline — that is a separate benchmark.

How QPC reports results — two lanes

This challenge is Lane A (Topology Certificate). Customer-facing Heron pilots are Lane B — see trilogy overview.

Lane A · Topology Certificate (this page)

Question: Is one coupled job measurably different from K separable jobs merged in software?

  • Metric: raw ICC · dual controls (separable + intracontext)
  • Reporting: pre-registered ladder · auditable IBM job IDs
  • Extension: MCGS physics-motivated instance on the same stack

Lane B · Industry pilots

Question: What business number does polycontextural execution produce?

  • Metric: domain decoder scores on quantum samples
  • Reporting: Cerrado · Traffic · Allstate on ibm_fez
  • See: Heron trilogy

ICC measures non-separability of coupled structure — not best portfolio or optimization score. L4 Cerrado merge is a structural control within Lane A; full 88Q industry scores are Lane B.

How we report results (pre-registered): Primary pass/fail uses raw IBM counts (not error-mitigated). Mitigated ICC is a secondary check. L5a: 3/3 pass raw, 2/3 pass mitigated (seed 11 ratio 2.5×). L5b: intracontext control — boundary ICC is depth-independent across transpiled depths ~82 (3-job separable), ~127 (one-job intracontext), and ~200–270 (coupled cross-context only).

Polycontextural power (June 2026): Optimized transjunction parameters on IBM Fez: raw ICC gaps 0.289 · 0.101 · 0.130 (seeds 7, 11, 42) vs separable < 0.02. Intracontext control (L5b): gaps 0.291 · 0.101 · 0.145 on IBM — boundary ICC tracks cross-context coupling topology, not depth alone (controls at ~82 and ~127 depth show ICC ≈ 0).

The question no vendor publishes a pass/fail answer to: when carbon, biodiversity, and social objectives live in separate IBM jobs, does classical merging recover the same joint structure as transjunctions in one circuit?

QPC says no — and registers the metrics before claiming victory.

What we measure

  1. ICC (inter-context correlation) — extra correlation at context boundaries in the coupled arm that independent jobs cannot match (L1–L3).
  2. Merge failure (L4) — three objective-specific jobs + vote merge vs one coupled decoder on the same Cerrado portfolio formula.
  3. L5 · Optimized + intracontext control — raw ICC vs 3-job separable (L5a); coupled vs one-job intracontext, depth-independent check (L5b).
  4. L6 · Public dare — invitation to falsify with job IDs (separable control only).

Ladder status

L1 · Toy separability ✓ PASS

3 synthetic contextures · 4096 shots · ibm_fez · seeds 7, 11, 42.

ICC gap: 0.052 – 0.106 (raw) · 3/3 seeds pass

Example coupled job: d8i3vdtv8cos73f549cg

L2 · Cerrado real data ✓ PASS

Goiás municipalities n=12, k=4 · Ribeiro 2026 open data · one coupled job vs three separable jobs.

ICC gap (raw): 0.088 · coupled job d8i42ic2upec739loi50

Portfolio: coupled decoder 3.224 vs greedy 3.042

L3 · Heron scale ✓ PASS (per-boundary)

n=28, k=7, morph_depth=2 · mean ICC gap below pilot bar, but biodiversity↔social boundary gap 0.098.

Job: d8i44rc2upec739loktg

L4 · Merge failure ✓ PASS

IBM Fez (n=12): coupled decoder beats mean separable strict score by +0.061.

Pilot (n=22, Aer): one QPC job 6.002 vs three-job classical merge 5.571 — delta +0.431.

IBM Fez (n=12): three-job separable mean 3.164 beats greedy 3.042 but loses to coupled decoder 3.224 (+0.061). Pilot (n=22, Aer): one QPC job beats three-job classical merge (+0.431) — merge underperforms one coupled submission on that instance.

L5 · Optimized + intracontext control ✓ PASS

5a — vs 3-job separable (raw primary): ICC gaps 0.289 · 0.101 · 0.130 · 3/3 raw, 2/3 mitigated (seed 11 mitigated ratio 2.5×). Coupled jobs: d90pcq6u9n7c73ali9lg · d90qatfqq29s738mhr00 · d90qv7mu9n7c73aljra0

5b — intracontext control, depth-independent (IBM Fez): one job, same coupling budget rewired within each context only (no cross-boundary transjunctions). Gaps 0.291 · 0.101 · 0.145 · ratios 22× · 9× · 31× · 3/3 pass.

Intracontext jobs: d91327fqq29s738msbo0 · d913fkfqq29s738mt460 · d913fnj57qjs73b5j150

Depth-independent evidence: boundary ICC ≈ 0 at transpiled depths ~82 (3-job separable), ~127 (intracontext one-job), and high only when coupling crosses contexts (~200–270). Logical pre-transpile 2Q count is 42 in both coupled and intracontext arms; post-routing depths differ — see results/jsc_l5_depth_matched_ibm_fez.json.

L6 · Public dare ✓ READY

Kit: jsc_l6_challenge_kit/. L5 complete — dare is open. Run separable control only; match coupled raw ICC (gap ≥ 0.05) without one coupled circuit.

Summary table

LevelMetricResultStatus
L1ICC gap (toy, 3 seeds)0.052 – 0.106Pass 3/3
L2ICC gap (Cerrado n=12)0.088Pass
L3Per-boundary ICC (n=28)0.098 (bio↔social)Pass
L4Coupled decoder − merge+0.061 (IBM) · +0.431 (pilot)Pass
L5aRaw ICC vs 3-job separable0.289 · 0.101 · 0.1303/3 raw · 2/3 mit
L5bRaw ICC vs intracontext (depth-independent)0.291 · 0.101 · 0.145Pass 3/3
L6Vendor reproduction dareKit readyOpen

What this is not

Honest scope: JSC proves joint quantum structure, not that classical computers cannot optimize the same spreadsheet.

Reproduce & falsify

Pre-registration: docs/QPC_JOINT_STRUCTURE_CHALLENGE.md · Machine bundle: results/qpc_jsc_bundle.json

.venv/bin/python3 vendor_benchmarks/qpc_separability/jsc_depth_matched_runner.py --mode aer
.venv/bin/python3 vendor_benchmarks/qpc_separability/jsc_depth_matched_runner.py --mode ibm --seeds 7,11,42
.venv/bin/python3 vendor_benchmarks/qpc_separability/qpc_jsc_verify.py --shots 2048
Full ICC methodology → Cerrado technical report → K=4 vs K=1 ablation → L6 challenge kit → Parameter sweep results →

L6 invitation: IBM, IonQ, Quantinuum, Google, Amazon Braket teams — run the separable control with our published instance parameters (jsc_l6_challenge_kit/l1_instances.json). If you reproduce our coupled raw ICC (gap ≥ 0.05) without one coupled circuit, QPC withdraws the architecture claim for that instance. Send reproducible job IDs to readytogo@quantumpolycontextural.ai.