Density-dependent cell-tracking errors and kinematic sensitivity in zebrafish: deletion and wrong-link stress tests

We measured density-related cell-tracking error and tested its downstream effect on kinematic separability defined by late-position labels. Sparse Biohub annotations cover 128,883 links in 199 crops from two embryos. In a base-output report using fixed detections and association-only leave-one-embryo-out evaluation, edge false negatives increased from 2.8% to 12.6% across density quintiles. Separate production and baseline evaluations attributed most of their density gradient to wrong links. We analysed two Zebrahub recordings with deletion scenarios, then repaired the statistical design using matched class counts, fixed time support and independent calibration and evaluation experiments. The new iAUC endpoint had null rejection fractions of 0/100, 0/100, with wide Monte Carlo uncertainty. Curve-midpoint tests frequently failed estimability checks and did not validate the earlier 10-frame timing threshold. Neighbour-constrained suffix swaps changed actual motion while preserving all positions and times; 20 fresh paired runs per recording gave median iAUC differences of -0.000654, 0.000175 against controls with identical event counts. These results describe conditional sensitivity of available trajectories. They do not identify biological commitment onset or establish equivalence under tracking error.