Synthetics

The study of finite representations of lawful systems — how laws are produced, transformed, and partially lost under changes of description.

The Premise

Statistical Lawhood Under Finite Description

Every quantitative theory is built by throwing away detail. The usual reading treats this as a practical concession: the world is large, the page is finite, so we coarse-grain and apologise. Synthetics treats the description itself as the physical object. A representation sorts the antecedent generator structure into three parts — null law, the background the description absorbs; visible cost, the complexity it retains and pays for; and closure residue, the structured way it fails.

Four axioms on scalar law-content then force a unique measure of lawhood — the KL divergence — and with it a definite accounting. Effective physics is classified as structured closure failure, and the thermodynamic shape that every coarse theory seems to acquire stops being a coincidence.

The synthetic decomposition A lawful world of many trajectories is forced through a finite aperture — the description — and sorted into three channels: null law absorbed to the background, visible cost retained and paid for, and closure residue as structured failure. a lawful world the description null law absorbed visible cost retained · KL closure residue structured failure
The Result Ladder

Working Drafts

Five drafts, presented foundation first. None is refereed yet; the derivations are in the PDFs.

5 working drafts /assets/docs/ not refereed
foundation first
  1. Synthetics: Statistical Lawhood Under Finite Description
    Working draft — the foundation
    Four axioms on scalar law-content force the KL divergence as the unique measure of lawhood, and the synthetic decomposition — null law, visible cost, closure residue — follows as bookkeeping rather than metaphor. Effective physics is classified as structured closure failure. Quantum mechanics appears as the rigid spectral fixed point of partition-dynamics obstruction.
    PDF the-law-of-laws.pdf
  2. The Cost of Complexity
    Working draft
    Representations are instantiated in matter, and matter has a price. Drawing a distinction — defining a macrostate boundary — carries a strictly non-zero thermodynamic cost, so choosing a coarser description is a physical act that incurs a physical bill. This is the ledger the rest of the programme spends against.
    PDF the-cost-of-complexity.pdf
  3. Equilibrium from the Inside Out: Exact Hamiltonian of Mean Force
    Working draft — the exact case
    The Hamiltonian of mean force is usually bought with a weak-coupling apology. Dropping the requirement of system–bath commutativity gives the closure exactly, non-perturbatively: a reduced equilibrium description that survives strong coupling. One place where the general programme meets an exact result.
    PDF HMF_AI.pdf
  4. Self-Consistent Loss Geometries and the Laws of Learning
    Working draft
    Learning happens in physical systems constrained by energy budgets. Three postulates force the loss function to be a local physical quantity rather than a global criterion imposed from outside — adaptation as a consequence of the physics, not an optimisation story laid on top of it.
    PDF self-consistent-loss-geometries.pdf
  5. The Laws of Learning: Dynamical Phases of Hierarchical Adaptive Systems and the Ultraepistemic Catastrophe
    Experimental draft (algorithmic research) — an active laboratory, not a final formalism

    Force many components to coordinate through one narrow channel and each part must update against a delayed reflection of the whole. Past a critical effective gain the stable configuration flips to an oscillating 2-cycle — the ultraepistemic catastrophe — and the energetic and operational loss surges.

    Hierarchy is the dampener. Sequential, modular layers attenuate the transmission of the macro-state and postpone the catastrophe to much larger populations, which sorts adaptive systems into flat, modular and inertial phases: the flat phase shatters on growth, the modular phase balances instability against the cost of depth, the inertial phase is safe and cannot learn. Layer normalisation inside a deep transformer is this attenuation — the same operation under the canonical normal form, formalised as a real-space Laplacian renormalisation group flow.

    PDF the-phases-of-hierarchy.pdf
The Engine

EigenEngine

The theory above is also a machine. EigenEngine compiles a corpus into a graded structure — the conceptric — and generates documents from it by reconstruction rather than storage. Its paper and talk were generated by the engine, and its page carries its own build record.

paper + talk, engine-generated /demo/ — the editor over a real corpus the build record is on the page
The engine → Open the demo →