Transjunction coupling — ON vs OFF

Matched ablation on a T03 cluster slice: 3×26Q = 78Q, same seeds and 1024 shots, small-world ring transjunctions enabled vs disabled. Primary metric: bridge-local inter-context correlation (ICC) on ring-linked qubit pairs, with QPC readout mitigation (qpc_noise_reducer).

Verdict

Both IBM Heron backends pass |Δ bridge ICC| > 0.05 (readout mitigated, June 2026).

Coupled runs show higher bridge ICC than decoupled — consistent with transjunction imprint on ring bridge qubits.

Full-width ICC (all 26 aligned qubit indices) stays near zero (~0.003 Δ) — expected: most pairs were never bridged. This page reports bridge qubits only.

Results by backend

Backend	Coupled ICC (mit)	Decoupled ICC (mit)	Δ ICC	Δ bridge ZZ	Coupled job	Decoupled job
ibm_fez	0.072	0.008	+0.064	+0.117	`d8fh39ralsvc7391oe30`	`d8fh3ro7jphs739m98n0`
ibm_marrakesh	0.095	0.026	+0.069	+0.156	`d8fi333alsvc7391pvag`	`d8fi34jalsvc7391pve0`

Verify jobs in the IBM Quantum console. Machine-readable summary: results/coupling_ablation_proof_summary.json

Proof figure — Marrakesh (primary)

Mean bridge ICC (readout mitigated): coupled vs decoupled

Coupled ON

0.095

Coupled OFF

0.026

Δ = +0.069 · threshold 0.05 shown as reference line in analysis JSON

Method (short)

Layout: 3 contexts × 26Q morphogrammatic brickwork (depth 8, seeds 7000–7002).
ON: small-world ring transjunctions (3 bridge qubits per adjacent pair).
OFF: identical context circuits, bridges removed.
Observable: ICC on physical bridge pairs (e.g. q25↔q26), not all aligned indices.
Noise layer: project to ≤16 bridge qubits → QPC readout mitigation → recompute ICC/ZZ.

Developer notes (MD) Noise Reducer report

Honest scope

This is a controlled coupling ablation, not a domain outcome claim (see QQ pilot for prediction ablation).
Ring wrap pair (ctx2↔ctx0) can show weaker ICC than adjacent pairs — see per-pair JSON.
Complements open-instance scale evidence (T01–T05) on the same Heron fleet.