“Integration is biography” had a null hypothesis to defeat — and the falsifier has now been run. The result — that ~22% of seeds reach HIGH $\intinfo$, that the category is set by the cumulative trajectory rather than initial conditions — is romantic, and it is also exactly what one would see if $\intinfo$ were a noisy quantity in a system with sensitive dependence on its evolutionary path. Under that null, the “trajectory” is just integrated noise, the “HIGH” seeds are the lucky tail of a stochastic process, and the slope that “cleanly separates categories” is fit to its own outcome. Because this program has already retracted one finding to a measurement artifact — the zeroed-buffer / 1D-energy-counter collapse (see above) — the standard of proof was set high. The explicit obligation was: re-measure the seeds from saved state, show that Φ replicates on an independent re-run, and show that it is predictable from trajectory features computed before the outcome is known.

Verdict: the biographical reading survives the noise null. The full 50-seed V32 protocol was re-run from scratch on independent hardware (an NVIDIA GB10 under JAX 0.10.1, versus the original run); analysis code is v32_replication.py, a pure-numpy re-implementation of the Φ measure runnable offline from the committed per-seed records. (1) Cross-run reproduction: per-seed late $\intinfo$ reproduces between the two independent runs at $r = 0.951$ (50/50 seeds, mean absolute difference $0.0055$); matched per-cycle Φ trajectories agree to $\sim 10^{-8}$, and the category distribution reproduces (HIGH 24% vs 22%, mean $\intinfo$ 0.086 vs 0.088). A noisy quantity with no stable latent cause would not reproduce per-seed across independent runs; this one does. (2) Degeneracy guard: every alive-bearing snapshot (334 of them; 16 extinction-moment snapshots correctly skipped) carries real signal — zero degenerate hidden states — so the zeroed-buffer artifact that sank the earlier analyses is demonstrably absent from this (V27/V32) path. (3) Predictability from pre-outcome features: late $\intinfo$ is forecast by the mean drought bounce ($r = 0.60$) but not by the first bounce alone ($r = -0.07$) — the category is built by the sustained pattern of recovery, the trajectory-dependence the biographical reading claims, not a single lucky event. The falsifier passed on all three counts; “integration is biography” has met its stated test against the measurement-noise null.