Danil Gorinevski

2papers

2 Papers

3.1DCApr 24
Mathematical Foundations for Peer-to-Peer Lattice Computation

Danil Gorinevski

Five propositions on synchronous peer-to-peer computation on a grid graph in $\mathbb{Z}^2$. \textbf{Proposition~1} gives three lower bounds: a transport-work bound $\sum_i a_i \ell_i \ge W_1(μ,ν)$, a completion-depth bound $D_{\min} \ge r_μ$ on the support radius, and a compressive-reduction edge bound $|E'| \ge \operatorname{St}_G(\operatorname{supp}(μ)\cup\{x_\star\})$. A negative result: for corner-sink dimension-order routing, the sink-trunk route-load functional has variance $Θ(f(1-f)P^2)$ under i.i.d.\ Bernoulli activation, refuting $O(fP^{3/2})$ concentration. \textbf{Proposition~2}: under an $αβγ$ collective-communication cost model and a sparse-activation model, the grid-to-cluster latency ratio improves monotonically as $f_{\mathrm{act}}$ shrinks whenever the cluster's fixed overhead dominates the grid's geometric constant. \textbf{Proposition~3}: a sufficient algebraic criterion for schedule-independent reduction semantics is that update rules decompose into a local map and an abelian-monoid merge, as a product-preserving functor from the Lawvere theory of commutative monoids into a hardware-state category. Under an idealised non-interfering scheduler this yields a wall-clock bound; the criterion is PCC-certifiable. \textbf{Proposition~4} bounds the conditional expected route length under i.i.d.\ site failure in the subcritical regime $δ< p_c^{\mathrm{site}}(\mathbb{Z}^2)$ by an additive detour term, via Aizenman--Barsky cluster-size decay and a boundary-length bound. \textbf{Proposition~5} augments the grid with $k$ long-range Watts--Strogatz shortcuts per node, collapsing typical shortest-path length from $Θ(\sqrt{P})$ to $O(\log P)$; percolation is placed in the mean-field universality class (rigorous for the $1$-D-ring base, conjectural for the $2$-D-grid base).

23.0SEApr 6
Nidus: Externalized Reasoning for AI-Assisted Engineering

Danil Gorinevski

We present Nidus, a governance runtime that mechanizes the V-model for AI-assisted software delivery. In the self-hosting deployment, three LLM families (Claude, Gemini, Codex) delivered a 100,000-line system under proof obligations verified against the current obligation set on every commit. The system governed its own construction. Engineering invariants - traced requirements, justified architecture, evidenced deliveries - cannot be reliably maintained as learned behavior; assurance requires enforcement by a mechanism external to the proposer. Nidus externalizes the engineering methodology into a decidable artifact verified on every mutation before persistence. Organizational standards compile into guidebooks - constraint libraries imported by governed projects and enforced by decidable evaluation. Four contributions: (1) recursive self-governance - the constraint surface constrains mutations to itself; (2) stigmergic coordination - friction from the surface routes agents without central control; (3) proximal spec reinforcement - the living artifact externalizes the engineering context that RL and long-chain reasoning try to internalize; the specification is the reward function, UNSAT verdicts shape behavior at inference time, no weight updates; (4) governance theater prevention - compliance evidence cannot be fabricated within the modeled mutation path. The constraint surface compounds: each obligation permanently eliminates a class of unengineered output. The artifact's development history is a formal development - every state satisfies all active obligations, and the obligation set grows monotonically.