A fixed covariance rule gave no clear gain in odor discrimination accuracy for 246 traced sister pairs from two mice

We tested whether joint response variation improves an odor readout from cells with a traced common glomerular input. The panel had 246 same-plane sister pairs in 38 glomeruli from two anesthetized mice. Each recording had 47 odors and four repeats. All six splits used two repeats for training and two for testing. A fixed linear rule retained half the fitted cross-cell covariance. Its comparison used each cell variance alone. We tested odor pairs within delivery modules. Equal-mouse accuracy was 57.438% for the variance rule and 57.440% for the covariance rule. The change was +0.0025 percentage points. A conditional group resampling range was -0.147 to +0.168 points. The two recording changes were -0.051 and +0.056 points. Full fitted covariance reduced mean accuracy by 0.199 points. Recombining held-out cell values changed the mean rule difference by 0.057 points, but this secondary effect depended on the response definition. Under independent zero-mean repeats, exact recombination from two test repeats retains half the expected cross-cell covariance. It does not create a zero-covariance test set. These results give no clear gain for this pooled linear rule. They leave larger groups, other readouts, fast timing and other mice untested. We preserve compact inputs, pinned source records, all scores and reference draws, code, validation, figures and repeat instructions.

olfactory bulb · sister cells · anatomical common input · odor discrimination · repeat covariance · calcium imaging · linear readout · held-out repeats · reproducible benchmark