# QPC Coupling Ablation — Transjunction ON vs OFF

**Experiment:** T03 cluster slice — **3 contexts × 26Q = 78Q**, matched seeds/shots, **transjunctions ON vs OFF**.

## Primary metric (v2)

| Layer | What it measures |
|-------|------------------|
| **Full-width ICC** | All 26 aligned qubit pairs — often **dilutes** signal (~0.003 Δ) |
| **Bridge-local ICC** | Ring transjunction pairs only (e.g. q25↔q26) |
| **+ QPC readout mitigation** | `qpc_noise_reducer` on ≤16 bridge qubits — **primary public metric** |

**Threshold:** |Δ mean bridge ICC| > **0.05** (ON vs OFF, readout mitigated).

## Verified results (1024 shots, June 2026)

| Backend | Coupled job | Decoupled job | Δ bridge ICC (mit) | Δ bridge ZZ (mit) |
|---------|-------------|---------------|--------------------|-------------------|
| **ibm_fez** | `d8fh39ralsvc7391oe30` | `d8fh3ro7jphs739m98n0` | **+0.064** | +0.117 |
| **ibm_marrakesh** | `d8fi333alsvc7391pvag` | `d8fi34jalsvc7391pve0` | **+0.069** | +0.156 |

Public page: `QPC_COUPLING_ABLATION.html`  
JSON: `results/coupling_ablation_proof_summary.json`

## Run new ablation

```bash
cd site_release_2025_11_15
unset QISKIT_IBM_TOKEN IBM_QUANTUM_TOKEN   # use ~/.ibm_quantum_token
export QISKIT_IBM_INSTANCE=open-instance
export QISKIT_IBM_PLANS_PREFERENCE=open

.venv/bin/python scripts/run_qpc_coupling_ablation.py --dry-run
./run_qpc_coupling_ablation.sh ibm_marrakesh 1024
```

Hardware runs now include **bridge_local** + readout mitigation in the JSON automatically (schema `qpc_coupling_ablation_v2`).

## Re-analyze existing jobs

```bash
.venv/bin/python scripts/analyze_coupling_noise_reduction.py \
  --ablation-json results/coupling_ablation_marrakesh_1024shots.json \
  -o results/coupling_marrakesh_bridge_mitigated.json

.venv/bin/python scripts/build_coupling_proof_summary.py
```

## Code

| Path | Role |
|------|------|
| `scripts/run_qpc_coupling_ablation.py` | Hardware CLI |
| `scripts/analyze_coupling_noise_reduction.py` | Post-hoc bridge + mit analysis |
| `scripts/build_coupling_proof_summary.py` | Merge Fez + Marrakesh summary |
| `qpc_orchestrator/coupling_ablation.py` | Runner (v2 schema) |
| `qpc_orchestrator/coupling_metrics.py` | Bridge pairs, ICC, readout mit |

## Honest scope

- Claim applies to **bridge qubit pairs** with **readout mitigation**, not full 78Q aligned ICC.
- Ring wrap pair (ctx2↔ctx0) can remain weaker than adjacent pairs — report per-pair tables.
- Complements **QQ pilot** (outcome ablation) and **open-instance scale** (T01–T05).