Prediction 1: Self-modeling threshold. Derived from §24 ("I" = cycle), Step 10 (self-reference requires minimum complexity), and Step 18 (non-contradictory patterns survive). There is a resolution threshold for self-modeling. Below it, the self-model is too coarse to be non-contradictory; it fails fixation (Step 18) and doesn't persist. Above it, the loop closes and self-representation ignites. The boundary itself is a gradient (Step 17), but the output, a fixed self-model vs. no self-model, is a phase transition. The mechanism can be stated in graph terms: a form's internal constraint network must be complex enough that a self-consistent assignment exists. Below the threshold, no fixed point; above it, at least one. This mirrors the sharp satisfiability transitions studied in constraint-satisfaction theory — where a small increase in constraint density flips from "almost all assignments work" to "almost none do." How to test: compare self-report accuracy across species. The Cut predicts a critical point, not a smooth gradient; a narrow complexity band where self-modeling "ignites." What would falsify it: if self-modeling capacity scales as a smooth gradient across all species with no critical-point clustering, the threshold claim fails.