exp-rule-multigen + exp-multigen-coupling-ablate established that signal_type chain progress→combined→concept is COUPLING-DEPENDENT (mirror falsified). Now characterise: how constrained is this developmental program across diverse initial conditions? Variance test: N=10 runs with diverse seed_sets, coupling=ALL. Three possible outcomes: - HIGH CANALIZATION: chain progress→combined→concept replicates in ≥7/10 runs. Strongly canalized developmental program. Like gastrulation in vertebrates.
- MODERATE CANALIZATION: chain replicates in 4-6/10. Multiple developmental basins; coupling produces structure but with non-trivial variance.
- LOW CANALIZATION (noise): no consistent chain. Coupling enables proposals but doesn't channel them. Lost-chain-on-noise rather than retrieval.
Pre-registered: - Operationalise "chain replicates" as: signal_type passes through ≥3 distinct values, including at least one of {combined, concept}.
- Track per-run: max_depth, n_accepted, signal_type trajectory, voting_threshold trajectory, rules_changed_count.