Antenna placement pilot — closed case

Status (June 2026): Not a public flagship. True parallel multi-contexture runs were attempted on IBM Heron; strict case-brief acceptance did not pass. Documented honestly as architecture learning. Active QPC public line: separability (ICC) and architecture program.

Leonardo multi-frequency antenna placement (mAPP) — combinatorial one-hot sites with exactly k active antennas. Reference: arXiv:2511.15566. QPC polycontextural runs were compared to penalty-QAOA and constraint-preserving baselines on IBM Fez.

Primary metric

Feasible-placement rate: fraction of shots satisfying one-hot per site and exactly k active antennas. Secondary: coverage minus interference cost on feasible samples only.

IBM Fez summary (1024 shots, June 2026)

Method	Feasible %	Job ID (representative)
Penalty QAOA baseline	17.3%	`d8gsbm5v8cos73f3lnu0`
Constraint-preserving baseline	15.9%	`d8gsghtv8cos73f3lu6g`
QPC polycontextural	4.5–22.7%	`d8gsl09e8nrc73bgfh40` … `d8hd1cs2upec739ktbd0`
Strict case-brief acceptance (v3)	0.0%	`d8hi471e8nrc73bh90dg` …

NISQ depth and decode policy strongly affect feasible rate. Circuit construction and mixer design are proprietary — not published on this site.

What we learned

Parallel multi-contexture execution on Heron is feasible (representative jobs above).
Strict collapse predicates without classical repair are hard at published NISQ depths.
Separability ICC tests carry the falsifiable architecture claim; antenna optimization does not.

Architecture program →