# QPC vendor replication catalog — real-world quantum tasks with public data

**Goal:** Find corporate/vendor quantum optimization (or application) projects where **problem definition + results** are published precisely enough to run the **same task** on **IBM Heron via QPC**, with a **comparable metric** and auditable job IDs.

**Rule (same as comparable benchmarks sheet):** comparable only if (1) numeric objective, (2) external baseline published, (3) reproducible on gate-model IBM with same metric.

---

## Your D-Wave PDF — important finding

**Link:** [Quantum Computing: The Key to Addressing Today’s Complex Business Problems](https://www.dwavequantum.com/media/sdylxhd0/quantum-computing-the-key-to-addressing-today-s-complex-business-problems.pdf)

| What it is | What it is **not** |
|------------|-------------------|
| Wakefield Research **survey** (400 business leaders, May 2025) commissioned by D-Wave | A quantum **execution report** |
| Opinion on logistics, ROI, AI vs quantum adoption | QUBO matrices, job IDs, instance files, or solver traces |
| Useful for **market narrative** (supply chain, routing, inventory) | **Not** sufficient to replicate a computation |

**Verdict:** ❌ **Cannot generate the same task from this PDF alone.** Use it only to pick *problem class* (e.g. vehicle routing, flow-shop). For data, use the reproducible sources below.

---

## Tier A — Replicate now (precise public data + IBM-feasible)

### A0. Cerrado carbon-credit portfolio — **BUILT** ✅

| Field | Detail |
|-------|--------|
| **Who** | Hugo José Ribeiro, Universidade Federal de Goiás |
| **Paper** | [arXiv:2602.09047](https://arxiv.org/abs/2602.09047) — QAOA + ZNE for Brazilian Cerrado municipality selection |
| **Code / data** | [github.com/hgribeirogeo/qaoa-carbon-cerrado](https://github.com/hgribeirogeo/qaoa-carbon-cerrado) |
| **Problem** | Select **k=28** from **n=88** Goiás municipalities; carbon + biodiversity + social synergies |
| **Metric** | Weighted **portfolio score** (same `evaluate()` as paper) |
| **Published baselines** | Greedy **44.42**; QAOA+ZNE mean **~58.47** (`ibm_fez`, `ibm_torino`) |
| **QPC build** | `vendor_benchmarks/cerrado/` — greedy verified **44.419**; Aer dry-run n=88; report `QPC_CERRADO_COMPARE.html` |
| **QPC angle** | One polycontextural submission (3 objective contextures + transjunction bridges) vs sequential 3-job hybrid vs single weighted QAOA |
| **Next** | `python3 vendor_benchmarks/cerrado/qpc_cerrado_compare.py --mode ibm --backend ibm_fez` |

---

### A0b. Volkswagen Beijing traffic-flow — **BUILT** ✅

| Field | Detail |
|-------|--------|
| **Who** | Volkswagen / D-Wave (Neukart et al. 2017) |
| **Paper** | [arXiv:1708.01625](https://arxiv.org/abs/1708.01625) — traffic flow on quantum annealer |
| **Data** | [T-Drive](https://www.microsoft.com/en-us/research/publication/t-drive-trajectory-data-sample/) + OSMnx |
| **Problem** | Assign 1 of 3 routes per taxi; minimize **congested-road count** |
| **Published baselines** | qbsolv **~40–80** vs random **~120–160** congested roads (418 cars) |
| **QPC build** | `vendor_benchmarks/traffic/` — pilot 16 cars / 48 vars on **ibm_fez** |
| **QPC result** | Decoder **2** congested roads = greedy; beats random mean **7.46**; job `d8gq6m42upec739k6pig` |
| **QPC angle** | 3 zone contextures + transjunctions in **one** submission vs qbsolv-style 3-zone sequential jobs |
| **Report** | `QPC_TRAFFIC_COMPARE.html` |

---

### A1. LR-QAOA Weighted MaxCut — **best next QPC demo** ⭐

| Field | Detail |
|-------|--------|
| **Who** | Cross-vendor study (IBM, IonQ, Quantinuum, IQM, Rigetti, OriginQ) |
| **Paper** | [arXiv:2502.06471](https://arxiv.org/abs/2502.06471) — *Evaluating the performance of quantum processing units at large width and depth* |
| **Code / data** | [github.com/alejomonbar/LR-QAOA-QPU-Benchmarking](https://github.com/alejomonbar/LR-QAOA-QPU-Benchmarking) (MIT) |
| **Problem** | **Weighted MaxCut (WMC)** — three topologies: 1D chain, **native layout**, fully connected |
| **Metric** | Approximation ratio **r** vs **random-sampler baseline** (pass/fail at 99.73% confidence) |
| **IBM reference** | **ibm_fez** already in paper: e.g. **r ≈ 0.771 at p=10** (156Q native, fractional gates); depth scaling to p=1000+ documented |
| **Quantinuum reference** | H2-1 passes **56Q fully connected, p=3**, 4620 ZZ gates (different topology than IBM native) |
| **Why good for QPC** | Same device you use; fixed protocol (no heavy classical optimizer loop); published curves to cite; natural **multi-context** extension: 3–4 WMC instances as labelled contexts on one Heron layout (risk / cost / carbon weights) with **same r metric** per context + ICC across contexts |
| **QPC angle** | “Same LR-QAOA metric as Jülich cross-vendor benchmark; QPC adds co-resident multi-scenario MaxCut on one submission.” |
| **Effort** | Medium — 1–2 weeks: port instance generator, run LR-QAOA baseline + QPC multi-context variant |

---

### A2. D-Wave NL-Hybrid benchmark (45 instances) — MaxCut / TSP / Knapsack

| Field | Detail |
|-------|--------|
| **Who** | D-Wave (Osaba & Miranda-Rodriguez) |
| **Paper** | [arXiv:2410.07980](https://arxiv.org/abs/2410.07980) — NL-Hybrid solver analysis |
| **Data** | **Mendeley Data** [10.17632/fbj4ycdn98.1](https://doi.org/10.17632/fbj4ycdn98.1) — all **45 instances** + results + generator scripts |
| **Problems** | 15× TSP, 15× Knapsack, 15× **Maximum Cut (MCP)** as QUBO/BQM |
| **Their stack** | NL-Hybrid, BQM-Hybrid, CQM-Hybrid, Advantage QPU |
| **Metric** | Best objective / cost found (per instance); sample sets in repository |
| **QPC on IBM** | Take **MCP QUBO** instances → QAOA or QPC-orchestrated QAOA on **ibm_fez**; compare **QUBO energy** (same Hamiltonian, not annealing time) |
| **Honest scope** | Annealer vs gate-model — compare **objective value**, not wall-clock or D-Wave-specific hybrid plumbing |
| **Effort** | Medium — start with **smallest MCP instances** (table in paper) |

---

### A3. IonQ BF-DCQO protein panel — **DONE** ✅

| Field | Detail |
|-------|--------|
| **Paper** | [arXiv:2604.26861](https://arxiv.org/abs/2604.26861) |
| **Metric** | Energy (lower better) |
| **QPC** | IDWKK −9.74 vs −8.70 on Fez — already published |

---

### A4. IonQ apps-benchmark (closed cases)

| Field | Detail |
|-------|--------|
| **Repo** | [github.com/ionq-publications/apps-benchmark](https://github.com/ionq-publications/apps-benchmark) |
| **What** | 13 application benchmarks (optimization, chemistry, ML, …); **BenchmarkCase** UUIDs, scoring metadata |
| **IBM path** | DIY backend for IBM Runtime; run closed optimization cases on **ibm_fez** |
| **License** | **CC-BY-NC-ND** — read-only use; check before commercial site claims |
| **Metric** | Framework-defined success criteria + time-to-solution |
| **Effort** | Low–medium for one closed case; high to certify full suite |

---

### A5. GIC Case 3 — polycontextural QAOA (internal, classical baseline)

| Field | Detail |
|-------|--------|
| **Status** | **READY** in repo (`tools/generate_case3_gic2026_report_a4.py`, CO₂/QAOA scripts) |
| **Problem** | 40Q BESS siting QAOA |
| **Metric** | QAOA objective **J** vs classical grid baseline |
| **Note** | Not a vendor demo — but same *comparable* pattern as A1/A2 |

---

## Tier B — Reproducible with extra work (data public, formulation heavy)

### B1. D-Wave flow-shop scheduling (Taillard)

| Field | Detail |
|-------|--------|
| **Doc** | [D-Wave FSS vignette](https://docs.dwavequantum.com/en/latest/industrial_optimization/vignette_fss.html) |
| **Data** | **Taillard FSS instances** (industry standard, 120 files, 20–500 jobs) — public |
| **D-Wave metric** | Median optimality **gap** vs best known, 150s time limit |
| **IBM/QPC** | Must encode FSS as QUBO/constraint model; hybrid classical loop; start with **smallest Taillard** (20 jobs, 5 machines) |
| **Matches D-Wave PDF theme** | ✅ Supply chain / manufacturing — **this** is the real computational counterpart to the survey PDF |

---

### B2. QA-MetaLearning QUBO corpus

| Field | Detail |
|-------|--------|
| **Repo** | [github.com/qcpolimi/QA-MetaLearning](https://github.com/qcpolimi/QA-MetaLearning) |
| **Data** | QUBO instances, Pegasus embeddings, **QA/SA/TS solver sample CSVs**, optimal assignments (small instances) |
| **Use** | Pick one instance ID; run QPC QAOA on IBM; compare energy to **QA CSV best** and **known optimum** |

---

### B3. Quantinuum H2 QAOA benchmark folder

| Field | Detail |
|-------|--------|
| **Repo** | [github.com/Quantinuum/quantinuum-hardware-h2-benchmark](https://github.com/Quantinuum/quantinuum-hardware-h2-benchmark) — `/qaoa` |
| **Data** | JSON: circuits submitted + observed outputs |
| **Caveat** | Trapped-ion native; replay on IBM needs **recompilation** — compare **problem Hamiltonian expectation**, not identical circuit depth |

---

## Tier C — Not reproducible from public data (avoid as “same task”)

| Source | Why skip |
|--------|----------|
| D-Wave business survey PDF (your link) | No instances |
| D-Wave PoQW blockchain Leap demo | Different subject (ledger hash consensus) |
| Many enterprise “case studies” (Harel, etc.) | Narrative only; no open QUBO |
| arXiv:2504.08843 portfolio QUBO (D-Wave hybrid) | “Data upon reasonable request” |
| Vendor white papers without instance UUIDs | Aggregate charts only |

---

## Recommended QPC demonstration program (order)

| Priority | Task | External anchor | QPC story |
|----------|------|-----------------|-----------|
| **1** | **LR-QAOA WMC native layout, ibm_fez, N=156, p=10** | arXiv:2502.06471 Fig 3 | Same **r** as IBM paper; add **K=3 multi-context** WMC (scenarios) in one orchestrated run |
| **2** | **D-Wave MCP instance #1 from Mendeley** | arXiv:2410.07980 Table 2 | Same **QUBO cost**; QPC vs published NL-Hybrid sample best |
| **3** | **Taillard FSS smallest instance** | D-Wave FSS vignette | Supply-chain narrative from D-Wave PDF + real scheduling data |
| **4** | **IonQ apps-benchmark one closed opt case** | IonQ white paper | “Same benchmark case UUID on IBM via QPC backend” |

---

## Multi-context QPC design pattern (for A1 / A2 / B1)

For any Tier A QUBO task:

1. **Context 1** — cost weights (finance)  
2. **Context 2** — carbon / ESG weights (CO₂)  
3. **Context 3** — risk / robustness weights (supply shock)  
4. **Transjunction** — shared decision variables across contexts  
5. **Metrics:** per-context objective **J** or MaxCut **r**; plus **ICC** between context outputs (you already publish this methodology)

Compare to **single-context QAOA** at same qubit budget (same ablation logic as QQ pilot).

---

## Next step (implementation)

If you approve **Priority 1 (LR-QAOA WMC)**:

1. Clone `LR-QAOA-QPU-Benchmarking` into `site_release_2025_11_15/vendor_benchmarks/lr_qaoa_wmc/`
2. Script `qpc_lr_qaoa_wmc_fez.py` — reproduce **5Q and 156Q native** baseline r
3. Script `qpc_lr_qaoa_wmc_multicontext.py` — 3 contexts on one Fez layout
4. Public page `QPC_LR_QAOA_WMC_FEZ.html` — r vs paper + job IDs
5. Update `QPC_COMPARABLE_BENCHMARKS.html`

Say **“build A1”** to start implementation.

---

## Quick reference links

| Vendor | Best public artifact |
|--------|---------------------|
| **D-Wave (technical)** | [10.17632/fbj4ycdn98.1](https://doi.org/10.17632/fbj4ycdn98.1), [FSS vignette](https://docs.dwavequantum.com/en/latest/industrial_optimization/vignette_fss.html) |
| **IonQ** | [apps-benchmark](https://github.com/ionq-publications/apps-benchmark), [BF-DCQO arXiv:2604.26861](https://arxiv.org/abs/2604.26861) |
| **Quantinuum** | [H2 benchmark repo](https://github.com/Quantinuum/quantinuum-hardware-h2-benchmark), LR-QAOA FC 56Q in [2502.06471](https://arxiv.org/abs/2502.06471) |
| **IBM (in literature)** | ibm_fez curves in LR-QAOA paper — your Open-plan access |

*Note: “Continuum” in your message is treated as **Quantinuum** (trapped-ion).*