Readout noise set a threshold for transient information gain in three positive linear connectome models

We tested when transient signal growth improves ideal decoding of a pulse in three positive linear graph models. Stationary internal noise alone made information decrease. With added isotropic readout noise, a matrix test gave an exact threshold for any transient information gain. At recurrent gain 0.999, the threshold readout variance was 94.54, 0.2432, and 8.272 times the mean internal variance in selected worm, fly, and Allen-based mouse models. At twice each threshold, the best time-grid information ratio was less than 1.0021. Lower gains often gave no finite threshold. A proof, covariance checks and a closed two-cell example support the test. The graphs differ in scale and construction. No activity data or noise levels were fitted. These are model results near instability, not evidence of species differences or measured decoding gains. External graph records remain outside the package. Code, aggregate scores, hashes and retrieval instructions support reproduction.

connectome · transient amplification · Fisher information · readout noise · Allen Institute · linear dynamics