v2.17.0 - Wormhole N-scaling check: N=8 vs N=12 (Experiment 15)

Tests whether the sign-dependent instance variance from Experiments 10/11 persists, worsens, or shrinks at a larger Majorana count -- the one structural lever never pulled until now.

The exact backend is infeasible at N=12 (dim^3 diagonalization cost, ~4096x slower than N=8). Both N=8 and N=12 are evaluated here via the Trotterized gate-circuit backend (~19s/call at N=12, measured directly) for a clean, backend-matched comparison, rather than the exact-backend N=8 numbers used in Experiments 10/11 (mixing backends would confound N-scaling with a separate, already-quantified backend effect from Experiment 9).

Scope: n=6 instances per N (not n=100, infeasible at this cost), K_TERMS fixed at 10. The paper's own 34/11 selection criterion has no exact match at N=12 (verified: 0 of 3000 candidates) -- N=12 instances are selected by closest achievable match instead.

Result: wrong-sign rate identical (2/6 at both N, too small a sample to trust as a rate), but mean |delta| drops from 0.00765 (N=8) to 0.00034 (N=12) -- roughly a 22x reduction, present in every N=12 instance individually. Consistent with the signal weakening toward a thermodynamic limit, but equally consistent with the paper's fixed default parameters becoming increasingly sub-optimal at larger N -- this first look cannot distinguish the two explanations.

Full write-up: docs/wormhole_syk_teleportation.md#experiment-15-n-scaling-check-n8-vs-n12-run-2026-08-07