Experiments

V19: Bottleneck Furnace

Period: 2026-02-18. Substrate: V18 (boundary-dependent Lenia).

Question: Is the Bottleneck Furnace effect due to selection (culling low- $\intinfo$ patterns) or creation (the bottleneck triggering developmental changes)?

Design: Three phases — standard evolution (10 cycles), fork into BOTTLENECK / GRADUAL / CONTROL (10 cycles), then novel extreme drought with frozen parameters (5 cycles). Statistical test: $\text{novel\_robustness} \sim \phi_{\text{base}} + \text{is\_bottleneck} + \text{is\_gradual}$ .

Seed	Verdict	$\beta_{\text{bottleneck}}$	$p$	$R^2$	Mean rob A vs C
42	CREATION	+0.704	<0.0001	0.30	1.116 vs 1.029
123	Design artifact*	-0.516	<0.0001	0.32	1.016 vs 1.010
7	CREATION	+0.080	0.011	0.29	1.019 vs 0.957

*Seed 123 reversal: fixed 8%-regen stress failed to create actual bottleneck mortality; conditions were effectively swapped.

The Bottleneck Furnace is generative, not selective. Patterns that evolved through near-extinction show significantly higher novel-stress generalization that cannot be explained by pre-existing $\intinfo$ . The furnace forges integration capacity — it does not merely reveal it. Raw BOTTLENECK ≥ CONTROL in all 3 seeds.

Source code

v19_experiment.py — Three-phase experiment design
v19_gpu_run.py — GPU runner (Lambda Labs GH200)