We are still at no artificial-life claim. The last wave of work did not produce a life claim. It produced something arguably more important at this stage: a stronger falsifier around our substrate evidence. The honest headline is that we are not claiming artificial life, we are building the evidence machinery needed to know when such a claim would be real.
Still stuck at the same gate
The substrate still fails the strict open-ended-evolution gate, on the same two blockers: novelty plateaus, and complexity plateaus. The encouraging part is that the supporting rails are increasingly clean. Recent full-scale runs pass horizon, ecology persistence, behavioural change, and the falsifier checks, but they do not yet show sustained late novelty or complexity accumulation.
The falsifier that caught us
We built a richer challenge, where the organism has to handle composite resource targets rather than a single channel. At full scale it produced about 5,100 lineage rows and 33 live-composite advances, and the artifact verified cleanly. It still failed the gate on both novelty and complexity plateaus.
Then we built the control that mattered: a matched-random burst control. It matched the structural burst volume of the stall-coupled condition exactly, 1,182 burst children, but it fired that churn at random points rather than on live stall state.
The result was sharp. The stall-coupled condition gave 41 live-contrast advances. The matched-random control gave 42. Both failed the gate, both plateaued on complexity, and the control verified cleanly with no claim promoted.
So burst timing is not yet a defensible adaptive signal. Random churn, matched for volume, reproduced or slightly exceeded our headline result. The falsifier did its job: it stopped us over-reading a tempting number. That is the project getting more honest, not more theatrical.
What still holds, and what is still missing
The earlier result holds. A stateful drive raised expressible behavioural dimensionality from roughly 2 to about 5, which showed the substrate can express richer behaviour. But later runs showed that selectability and sustained novelty are the harder missing pieces. In plain English: the organism can express more than the ecology can yet select and retain.
We are not claiming life, artificial life, open-ended evolution, autonomy, agency, consciousness, or self-improvement. The standing blockers are sustained late novelty, sustained complexity accumulation, long-horizon survival under pressure, open-ended reproduction and descent with heritable variation, and independent model-family replication at claim-grade strength.
The next direction
The next experiment makes the resource challenge itself evolvable. Instead of fixed channels or burst timing, it tests whether lineages can pay for, inherit, and later exploit changes in the challenge itself. The question becomes whether history-made resource mappings can create retained late novelty under descent, while shuffled, removed, free, frozen-replay, and matched-random controls all break the signal. Packet-first, no claim, verifier-backed, and control-heavy.
That is the next honest step toward a claim that could survive scrutiny. We’ll post the next update when there’s something measured to report.