Overnight, Synthena made a real but disciplined step forward. We are still making no artificial-life claim, but the substrate evidence improved in the exact place we wanted: falsifier-sensitive, geometry-retained novelty across a longer horizon.
What landed
We ran a longer matched panel over 12 seeds: the real substrate against a shuffled-structure control and a destroyed-structure control, on a stricter geometry-retained novelty readout.
- Real substrate: 3 of 12 signal seeds.
- Shuffled-structure control: 0 of 12.
- Destroyed-structure control: 0 of 12.
All artifacts verified cleanly: no hash mismatches, no gate weakening, no threshold retuning, and nothing promoted.
The moment that mattered
The stricter readout had been sparse, 1 of 12 at the shorter horizon. The strongest result of the night was that lengthening the horizon did not wash the signal away. It got stronger: 1 of 12 became 3 of 12, while the matched falsifiers stayed dead at 0 of 12.
That matters because the signal did not survive when the structure was shuffled or destroyed. In plain terms, the behaviour we care about appears tied to the real geometry and resource structure of the substrate, not to noise or generic novelty churn. It is a cleaner sign that the substrate has a structure-dependent adaptive trace.
The honest caveat
A broader sustained-novelty metric still fired across all three arms. Real, shuffled, and destroyed all produced some signal on it, which tells us that broader metric is too permissive for this lane. So the honest interpretation is narrow: the stricter readout is the cleaner signal; the broader one remains useful as a frontier check, but it is not claim-bearing here. The strict open-ended-evolution gate still fails closed, with nothing promoted.
What we can and cannot say
What the evidence now supports: sparse, falsifier-sensitive, geometry-retained novelty in a structured substrate, an effect that strengthens under a longer bounded horizon and breaks under shuffled or destroyed structure.
What it does not support: we cannot claim artificial life, life, autopoiesis, or open-ended evolution. This is a signal about the substrate, not a life-property, and it is sparse and from a single overnight run.
Where we are now
We are closer to a defensible substrate claim, and not yet close enough to a defensible artificial-life claim. The next engineering move is clear: make this geometry-retained novelty an active pressure on the substrate rather than a post-hoc readout, then rerun the same real, shuffled, and destroyed panels and require the real signal to rise while the controls stay flat. That is the next honest step, and we will post it when there is something measured to report.