Clearance capacity sets the seed required for self-sustaining TDP-43 aggregation

TDP-43 represses its own transcript through its 3'UTR. When the protein leaves the nucleus that repression lifts, so the cell makes more, and the new protein partitions to the cytoplasm where the aggregate is: the negative feedback inverts into positive feedback. The loop and a conceptual threshold have been described, but no formal model exists, so the threshold has never predicted anything. We build one, constrain it against clinical genetics and against a published time course, and report three results. A three-variable model is bistable when aggregate clearance is capacity-limited rather than first-order - a change forced by a quantitative contradiction, not chosen, since first-order clearance caps the elongation rate constant six to eight orders of magnitude below a measured amyloid. The corrected model survives a clinical test the first version failed, and makes three results. First, clearance capacity is the dominant lever on the threshold: halving it lowers the seed required for self-sustaining aggregation by about 36 %, close to linearly. That the dominant lever is clearance was not built in, and it names the gene class ALS keeps implicating - OPTN, TBK1, SQSTM1, UBQLN2, VCP. Second, inverting a model failure against patient phenotypes gives a quantitative constraint on a cell: a motor neuron must clear aggregates at more than about twice the rate templated growth would consume them, or half-dosage carriers would be born ill rather than developing late-onset disease. Third, reanalysis of a published charge-variant time course shows the conversion hazard is constant in time (0.0047 h-1, no estimate more than 6 % from the mean over a four-fold range of time), that no construct shows a rising hazard, and that the reported linear dependence of conversion on net charge is a property of the readout rather than the mechanism. An extension of our own - a fixed susceptible subpopulation - was built, tested against held-out data, and rejected. We give four experiments that would falsify the framework.