July 2026 · Pre-registered · Four points

Multi-Contexture Gauge-Sector (MCGS)

Gyawali et al. motivate gauge-sector physics on Willow (arXiv:2410.06557). QPC runs a pre-registered architecture instance on the same ICC stack as the Joint Structure Challenge — not a Willow circuit re-run.

Plain English

What this test is
MCGS = Multi-Contexture Gauge-Sector: three quantum contextures (even / odd / bridge sectors) coupled in one IBM Fez job, with weight priors motivated by lattice-gauge physics.
Why we did it
To show the JSC ICC signature extends to a physics-motivated instance — and survives dephasing on real hardware with the same pre-registered bar.
What QPC computed
4096 shots · coupled vs 3-job separable vs intracontext control · Aer coupling sweep before Fez · four pre-registered points (two IBM, one ED toy, one analytic).
Headline result
July 2026 · 3/3 raw pass · gaps vs separable control 0.349 · 0.243 · 0.131 · ratios vs intracontext control 28× · 26× · 8× (seed 42 smallest margin on either control, still clears 3× bar).
What we claim — and do not
Claim: architecture ICC clears on Fez for this instance. Do not claim: Willow parity, supremacy, or “classical cannot simulate localization.”

Same occurrence, different instrument

What QPC delivers — forward framing

We claim: the superposition-versus-sampling occurrence that motivates disorder-free localization in lattice gauge theory is native to polycontextural architecture — even / odd / bridge gauge sectors mapped to three contextures in one IBM Fez job, with cross-boundary ICC on raw measurement counts (3/3 pass, gaps 0.349 · 0.243 · 0.131, public job IDs). Sector structure is topology, not a custom Hamiltonian program: the same logical fact Google’s team proved in physics on Willow, QPC makes executable and auditable on IBM Heron in one pre-registered protocol — in some hours of work.

Gyawali et al. (arXiv:2410.06557 · Science 2026) showed in fundamental physics that gauge-sector superposition ≠ incoherent disorder sampling. MCGS shows that distinction on commercial gate hardware: coupled polycontextural execution produces ICC that separable multi-job and intracontext controls cannot reproduce — coupling-borne, not depth-borne, no decoder, no mitigation stack.

Customer takeaway: QPC does not wait for a bespoke lattice-gauge circuit. It encodes sector splits as contextures + transjunctions and proves the architecture signature where buyers can verify it — IBM job IDs, raw counts, pre-registered bar.

What QPC proves on IBM FezHonest scope (not this page)
Coherent multi-sector structure vs factorized execution (ICC witness) Willow LGT circuit parity
One coupled job beats separable 3-job + intracontext controls (28× · 26× · 8×) Error-mitigated gauge polarization time series
Sector topology as first-class QPC encoding — fast path to auditable hardware proof Classical intractability or supremacy claims

Scope note: Point 1 is ED sanity; Point 4 is analytic motivation. This is Lane A architecture evidence, not a replacement for Google’s full experimental physics program.

How QPC reports results — two lanes

This page is Lane A (Topology Certificate). Industry pilots with customer scores live on Heron trilogy (Lane B).

Lane A · Topology Certificate (this page)

Question: Does coupled polycontextural wiring show up in raw hardware?

  • Metric: ICC vs separable multi-job and intracontext controls
  • Reporting: raw IBM counts · pre-registered bar · job IDs in tables below
  • This instance: MCGS gauge-sector weights on the JSC ICC stack

Lane B · Industry pilots

Question: What business outcome does polycontextural execution deliver?

  • Metric: portfolio · congestion · premium via declared problem decoder
  • Reporting: one coupled Heron job · customer meaning
  • See: Cerrado · Traffic · Allstate

IBM Fez result (July 2026): Raw ICC gaps vs 3-job separable control 0.349 · 0.243 · 0.131 (seeds 7, 11, 42) · 3/3 pass. Against the intracontext control, the same coupled results separate 28× · 26× · 8× (coupled ICC ÷ intracontext ICC). Every coupled result clears 3× against both controls — the separation tracks cross-context coupling, not any single baseline.

The signal is coupling-borne, not depth-borne, and two controls bracket the depth confound from both sides: the intracontext control runs at its own depth (~127–131) and sits at floor; the separable multi-job control is itself deep yet also stays at floor. If circuit depth alone produced ICC, the deep separable control would show it regardless of wiring — it does not. Only cross-context transjunction coupling raises ICC above floor.

No mitigation. No decoder. No classical solver in the loop.

The separation you see is in the raw quantum measurement counts. Points 2 and 3 pass on unprocessed IBM Fez output — no readout mitigation, no problem-aware decoder, no post-selection. The inter-context correlation is carried by the polycontextural circuit topology itself and is auditable from the public job IDs. Most published near-term quantum optimization results reach their headline figure through a decoder or an error-mitigation stack; this result does not — scoped to ICC architecture witness only, not to QPC industry optimization pilots.

How we report (pre-registered): Primary pass/fail uses raw IBM counts. Same JSC bar: coupled ICC ≥ 0.02, gap ≥ 0.05, ratio ≥ 3× vs each control. Hardware arm = soft-objective QPC circuits with sector-motivated weights — not Trotterized LGT dynamics. Point 4 (Null B) is an analytic disorder-average argument; Points 2–3 are the IBM headline.

Four pre-registered points

Two independent controls, two ratios. Each coupled result is tested against both a separable multi-job control (Point 2) and an intracontext control (Point 3). These are different baselines, so a given seed has two ratios — e.g. seed 11 separates 18× from the separable control and 26× from the intracontext control; seed 42 shows 10× vs separable and vs intracontext (higher intracontext floor). A pass requires clearing 3× against both.

Point 1 · Localization sanity (ED) ✓ PASS

1D Z₂-inspired patch · gauge-sector superposition over 8 physical states · confinement ξ∞ = 1.29 < bar 2.5.

ED reference only — not the hardware headline. Persistence margin weak at L=5 (finite-size); confinement is the primary ED bar.

Point 2 · ICC vs separable (IBM Fez) ✓ PASS 3/3

Three contextures with even / odd / bridge weight priors (motivated by gauge-sector split) · 4096 shots · couple_scale=0.8.

SeedRaw ICC gapSeparable ICCRatio vs separable control (≥3×)Coupled job
70.3490.01132×d93up0fu62ks7395c3k0
110.2430.01418×d93vk1dgc6cc73feg13g
420.1310.01510×d93vllvu62ks7395d7og

Aer pre-check: 3/3 · mean gap 0.458 at couple_scale=0.8.

Pre-registered Aer coupling sweep (before Fez): couple_scale 0.8 → 0.458 · 1.0 → 0.441 · 1.2 → 0.357 (mean ICC gap vs separable, 3/3 pass at each point). IBM ran at sweep-selected best (0.8) — not tuned after hardware.

Point 3 · Null A — intracontext control ✓ PASS 3/3

One job; coupling budget rewired within each context only (no cross-boundary transjunctions).

SeedIntracontext ICCRatio vs intracontext control (≥3×)Intracontext job
70.01328×d93up0vu62ks7395c3kg
110.01026×d93vk1kql68s73c8tnm0
420.019d93vlmfu62ks7395d7pg

Seed 42 shows the smallest margin (8×), driven by a higher intracontext floor (0.019 vs 0.010–0.013); it still clears the pre-registered 3× bar. Reported explicitly rather than absorbed into a range.

Point 4 · Null B — disorder-average ✓ PASS (analytic)

Not an IBM measurement. Analytic: sector probes a(s)=(-1)^s, b(s)=1 ⇒ Covs(a,b) = 0 — the connected term Γ in a coherent superposition is inaccessible to incoherent disorder sampling. This formalizes the motivation from Gyawali et al.; Points 2–3 test whether QPC’s ICC signature clears on hardware.

Summary

PointMetricIBM FezStatus
1ED confinement ξ∞1.29 < 2.5Pass (ED)
2Raw ICC gap vs separable0.349 · 0.243 · 0.131Pass 3/3
3Raw ICC vs intracontext28× · 26× · 8×Pass 3/3
4Null B Covs=0Analytic (ED note)Pass (theory)

Overall (pre-registered): IBM Points 2–3 3/3 pass raw · Point 1 ED confinement pass · Point 4 analytic pass · bundle results/qpc_mcgs_bundle.json

What this is not

Reproduce

Pre-registration: docs/QPC_MCGS_PREREGISTRATION.md · MCQST outreach draft: docs/QPC_MCGS_MCQST_OUTREACH.md

.venv/bin/python3 vendor_benchmarks/qpc_mcgs/mcgs_icc_runner.py --mode ed
.venv/bin/python3 vendor_benchmarks/qpc_mcgs/mcgs_icc_runner.py --mode aer-sweep
.venv/bin/python3 vendor_benchmarks/qpc_mcgs/mcgs_icc_runner.py --mode ibm --backend ibm_fez --seeds 7,11,42
.venv/bin/python3 vendor_benchmarks/qpc_mcgs/mcgs_bundle.py
Joint Structure Challenge → ICC methodology → Machine bundle JSON → IBM Fez results JSON → Scripts README →