Research · Index
Papers, essays, and instrument releases from SN-X residents and visiting faculty. We publish what we learn — including the parts that are mid-progress.
Self-organizing criticality is often treated as a fragile phenomenon requiring precise tuning of a control parameter. We show that in sufficiently heterogeneous networks, critical-like scaling emerges robustly across a finite region of parameter space — and propose a practical detector that does not require direct access to the underlying dynamics.
We study the intrinsic dimension and curvature of learned representations across 47 pre-trained models. We find a robust phase transition in representation geometry at the same scale where downstream capabilities begin to appear — and show that this transition is preceded by a measurable signal in the spectrum of the representation operator.
We recast developmental pattern formation as approximate Bayesian inference over a generative model of tissue. The resulting dynamics reproduce classical Turing patterns, recover experimentally observed stripe-spot transitions in zebrafish, and predict a previously unreported bistable regime in chick feather primordia.
An essay review: we argue that adaptive immunity is best understood as a continual, embodied reinforcement-learning process operating on a population of B and T cells, with somatic hypermutation playing the role of a parameterised policy and affinity maturation as policy-gradient ascent.
We release the first three SN-X Instruments: a criticality detector for time-series, a large benchmark for emergent communication in multi-agent simulations, and a dataset of regulatory networks reconstructed from single-cell data.
We formalize a price-mediated exchange market as a distributed message-passing algorithm computing a global price vector, and prove convergence under assumptions that match the empirically observed structure of limit-order books. The theory predicts a previously unreported phase transition between two qualitatively different price-discovery regimes.
Classical causal identification assumes the analyst can enumerate the relevant variables. We relax this assumption: when the variable set itself is uncertain, what can be identified? We give necessary and sufficient conditions, and demonstrate them on a climate-policy case study where the standard approach fails.