In one Area X GPi-STN export, local controls reduced self return excess but did not show a clear stability gain in a normalized model without delays

We tested self return and feedback stability in one adult zebra finch Area X export. The selected GPi-STN loop had 43 GPi cells, 71 STN cells and 22,158 contacts. We normalized contact-area weights by each target input sum. Mean two-step self return was 6.89%. Its exact reference mean was 2.33% under global profile exchange, 5.94% in 80 micrometre cubes and 6.76% in 40 micrometre cubes. The planned 80 micrometre trace test gave an upper-tail score of 0.0001 in 9,999 draws. In an isolated linear model without delays, the first instability gain was 35.83. It was inside the local 95% reference range of 14.84 to 38.78, with an upper-tail score of 0.057 in 999 draws. A stricter cell-label cohort gave stronger stability evidence. A four-cell example proves that identical self return at every cell can coexist with different instability thresholds. The base cohort therefore had excess self return, but no clear local stability advantage in this normalized model. These are conditional graph and model results, not measured neural function. The package includes new aggregates, source hashes, code, figures, a proof and a full exact repeat.

Area X · songbird · connectome · feedback · self return · linear stability · spatial controls · GPi · STN · reproducibility