# QPC Traffic-Flow Compare Task (Volkswagen / D-Wave 2017)

## External reference

- **Paper:** Neukart et al., *Traffic flow optimization using a quantum annealer*, [arXiv:1708.01625](https://arxiv.org/abs/1708.01625)
- **Data:** [T-Drive](https://www.microsoft.com/en-us/research/publication/t-drive-trajectory-data-sample/) + OpenStreetMap (OSMnx)
- **Original hardware:** D-Wave 2X via **qbsolv** hybrid (1,254 variables did not fit one annealer pass)
- **Metric:** **Congested-road count** (segment used in >10 routes on full 418-car instance)

## What QPC proves

Same public problem class, **reconstructed instance**, **gate-model IBM** execution:

| Workflow | IBM jobs | QPC role |
|----------|----------|----------|
| qbsolv-style partition | Many (classical loop) | **Anti-pattern** we replace |
| Sequential 3-zone QAOA | 3 | Zone partition without transjunction |
| Single weighted QAOA | 1 | Ribeiro-style baseline |
| **QPC polycontextural** | **1** | 3 geographic contextures + transjunctions on shared route variables |

## Pilot instance (IBM Fez, June 2026)

- **16 cars**, **48 binary variables**, synthetic 6×6 grid with shared arterial bottleneck (faithful QUBO recipe; not bit-identical to VW Fig. 3 matrix)
- Scaled congestion threshold (paper uses 10 @ 418 cars)

## Commands

```bash
cd site_release_2025_11_15
.venv/bin/python vendor_benchmarks/traffic/test_pipeline.py
.venv/bin/python vendor_benchmarks/traffic/qpc_traffic_compare.py --mode verify --n-cars 16 --grid 6
.venv/bin/python vendor_benchmarks/traffic/qpc_traffic_compare.py --mode dry-run --n-cars 16 --grid 6 --shots 4096

unset QISKIT_IBM_TOKEN IBM_QUANTUM_TOKEN
export QISKIT_IBM_DISABLE_INSTANCE_CRN_FILE=1 QISKIT_IBM_INSTANCE=open-instance
.venv/bin/python vendor_benchmarks/traffic/qpc_traffic_compare.py \
  --mode ibm --backend ibm_fez --n-cars 16 --grid 6 --shots 8192 \
  --ibm-jobs full -o results/traffic_compare_ibm_fez_16c_full.json
```

## Real Beijing data

```bash
pip install osmnx networkx pandas
# Download T-Drive CSVs into taxi_log/
python3 -c "
from vendor_benchmarks.traffic.traffic_pipeline import RealDataPipeline
rp = RealDataPipeline(tdrive_dir='taxi_log')
rp.load_graph()
rp.load_trajectories()  # when implemented
"
```

## Artifacts

- `results/traffic_compare_ibm_fez_16c.json` — QPC job `d8gq6m42upec739k6pig`
- `QPC_TRAFFIC_COMPARE.html` — public report