H9: The Evolving Network Hypothesis — Refinement Log

The 2026-07-27 code review withdrew sim03 (static-by-construction lattice) and sim05 (corrected to 2/6 coexistence, no longer a stall) as evidence, and corrected sim04's comparative result from 5-vs-4 cores to 3-vs-3 (the original run used a non-deterministic hash). Only sim04's finite-space exhaustion (510/510, both conditions) survives, and it doesn't test H9's claim. Vasas et al. (2012) remains the sole surviving support.

Topic: evolving reaction networks vs. fixed networks — evolvability via novel autocatalytic cores

The "one bit" limitation (original framing, Session 5)

A single viable core carries ~1 bit of heritable information (present/absent). Open-ended evolution requires unlimited heritable information. The evolving network extends the "adjacent possible" — each new core opens new reaction possibilities — but whether this produces true open-endedness or just limited multi-attractor dynamics is an open question.

Refinement (2026-07-27, post code review — two legs weakened, one survives)

H9's supporting evidence has thinned considerably.

  • Withdrawn: "sim03 negative." Its organization lattice is static by construction, so it cannot demonstrate that fixed networks fail to evolve.
  • Withdrawn: "sim05 shows unbounded space stalls differently." Corrected, sim05 gives 2/6 L2 coexistence — it does not show a stall.
  • Withdrawn: the comparative result. sim04 was recorded as showing the evolving network finding more autocatalytic cores (5 vs 4) — a "modest improvement" that was H9's only positive simulation evidence. Corrected, it is 3 vs 3, no difference, and the earlier figures came from a run that was not reproducible at all (catalysis was derived from Python's per-process-randomized hash()).
  • Survives: sim04's finite-space exhaustion. Both conditions saturate at 510/510 species and neither produces open-ended evolution. That is a structural bound and is unaffected by the determinism fix.

Net: H9's claim that evolving networks overcome the fixed-network stall currently has no supporting simulation evidence in this repo — the one measurement that appeared to support it has been retracted. Vasas et al. (2012) still supports it in the literature. H9 should be treated as untested rather than partially confirmed, and the test has in effect not yet been run under conditions that could answer it: sim04's finite space exhausts before the evolving mechanism has room to matter.