κυβερνήτης — the steersman
Cybernetics consulting. Lattice topology research. Open tools.
Every computation — neural networks, simulations, databases — defaults to cubic lattice topology: 6-connected grids. Each node talks to 6 neighbors along 3 axis-aligned directions. It is the geometry of spreadsheets, voxel grids, and attention heads. Nobody chose it. It was just there.
The FCC lattice gives each node 12 neighbors along 6 face-diagonal directions. Its Voronoi cell is the rhombic dodecahedron — a 14-vertex polyhedron that tessellates all of 3D space. Kepler proved it packs spheres optimally in 1611. Hales confirmed it in 2005. Under real-world heterogeneous conditions, it provides 6.1× the algebraic connectivity of the cube.
The cube is not the enemy — it is the skeleton. The rhombic dodecahedron contains the cube: its 8 trivalent vertices ARE the cube’s corners. The 6 tetravalent vertices are the bridges that convert cubic topology into space-filling FCC. TeLoRA adds a learnable bridge matrix between LoRA’s A and B projections. A cybernetic feedback mechanism (the Steersman) programs a chosen coupling topology into the bridge — the rhombic dodecahedron is the canonical target, not a discovery: 100% block-diagonal bridges, coupling ratios of order 10⁴–10⁵:1 (provisional pending controller re-verification), across four model families (1.1B–14B), at benchmark parity with standard LoRA. The mechanism is topology-agnostic, and we publish the null results that prove the geometry does not emerge from data on its own.
Two papers are out. Both were pre-registered before the data existed, audited adversarially by fresh-context agents, and published with their artifacts — protocol, code, raw logs, and the audit ledger, dispositions included. Where a prediction failed, the failure is in the paper.
A pre-registered bank of 480 LoRA adapters — two model families × six tasks × forty seeds — read out by weights alone. Raw adapter weights fail to identify the training task (0.0792 / 0.1375 against chance 0.1667); GL(r)-gauge-canonical features reach 1.0000. The obstacle is the parameterization, not the information. Weight-only merge-conflict prediction reaches group-aware AUC 0.995 / 0.962. Two of five pre-registered outcomes went against us — including a no-transfer prediction refuted by our own triviality control — and they are the credibility of the other three.
Read · PDF · Dataset · Audit ledger
When an agent compacts its own context, the format of the compacted state is chosen by convention rather than measurement. At matched token budgets, prose re-encoding corrupts 36.4% of numeric facts where typed state blocks corrupt 9.4% — paired exact McNemar p = 3.524×10−21, direction unanimous across three models. The loss happens at write time: typed blocks retain 99.8% of values across cascaded checkpoints, prose 65.2%. Never re-narrate a number when a quotable typed slot will do.
Read · PDF · Pre-registration
Cybernetics was born as the science of communication and control — the study of how systems regulate themselves through circular causation, not through command. TASUMER MAF applies this principle to AI systems architecture, organizational design, and computational infrastructure.
Communication and control. Named the field for the steersman who reads the wake and adjusts the rudder. Governance emerges from feedback, not force.
Requisite Variety. A regulator must have at least as much variety as the system it regulates. 12 connections absorb what 6 cannot.
Viable System Model. Organizational topology IS architecture topology. The structure of a viable system mirrors the structure of the environment it must absorb.
Ecology of Mind. The observer is part of the system. The map participates in the territory. The tool that measures connectivity is itself a connected system.
Autopoiesis. A system that produces the components that produce the system. Research produces evidence produces tools produces research.
Conversation Theory. Knowledge emerges through structured dialogue between participants. Human-AI collaboration as Paskian conversation.
Ethics of self-reference. “Act always so as to increase the number of choices.” Open methodology expands the space. Proprietary results fund the expansion.
Lattice topology benchmarks proving cubic networks leave 6.1× connectivity on the table. TeLoRA neural module with cybernetic Steersman feedback. Both released papers live here.
GitHub · PyPI · DemoAgent-driven adversarial paper audit framework, now a three-stage engine. Its Asset-1 ladder ran 7 rounds across 19 lenses: 117 confirmed findings, 92 fixed, every disposition on record.
GitHubResearch environment scaffold for Claude Code. Domain knowledge injection, multi-agent orchestration, epistemic charter enforcement.
GitHubThe autopoietic loop: each stage produces the next. The enterprise is the thesis made organizational.