With fixed means and independent errors of finite variance, repeat products give unbiased squared three-cell loop estimates

We give an estimator for the squared difference between forward and reverse three-cell response products. It uses at least two stimuli at each source in one animal. With fixed means and independent event errors of finite variance, its expectation is the true squared loop difference. Target errors within one stimulus can be correlated. In a reciprocal linear circuit with fixed source and readout gains, the true difference is zero. The zero condition does not need gain calibration. Exact enumeration, independent integration and 100,000-design simulations checked the estimator. A passive control on a 302-cell Cook electrical graph gave relative loop errors below 2.2e-15. We rebuilt weighted fluorescence responses from DANDI 001075 version 0.240930.1859. Ninety repeated loops in 15 WT C. elegans animals passed the recording checks. The main corrected mean was 1.7313e-8, with a descriptive whole-animal resample range from -0.4835e-8 to 5.8755e-8. Removal of one animal changed its sign. A fixed passive circuit with twofold gain changes and a separate finite-window example show limits of the physical inference. The recorded panel does not establish a failure of electrical reciprocity. The package preserves the proof, compact response data, anatomical facts, source hashes, all checks, code, figures and repeat instructions.

connectome · electrical synapses · reciprocal transfer · Caenorhabditis elegans · repeated stimulation · loop products · measurement error · structure and function