Durable Forms
An Essay on Load Bearing Representations
Working manuscript — Draft 0.21 · July 2026
Table of Contents
The Wasm homage, and how a naming gesture turned out to contain a thesis.
Part I — Inheritance
Plato’s theory of Forms recovered as the first projection architecture — durable templates, transient instantiations — together with its load-bearing flaw: the assumption that forms persist by their own nature. The book keeps the architecture and inverts the physics.
Susanne Langer’s symbols as vehicles of conception rather than proxies for objects, and Deleuze’s assemblages — heterogeneous parts held together by what they do, not what they essentially are — as the anti-essentialist corrective that keeps the book’s Platonism honest.
The scientific line: Shannon’s quantification-by-deferral of meaning, then Deutsch and Marletto’s counterfactual turn, in which information is constituted by which transformations are possible.
Chapter 2: The Atom of Distinction
How the bit made information substrate-independent and countable — the single yes/no distinction as the founding abstraction beneath compression, error correction, and every interoperable system since.
The deliberate deferral: semantics bracketed as “irrelevant to the engineering problem,” the most productive omission of twentieth-century science — and the debt it quietly accumulated in human heads, documentation, and tribal knowledge.
Entropy rehabilitated: not disorder but the richness of a shared, pre-existing possibility space — the codebook that must exist before any message can mean anything.
Transmission and retention contrasted point by point: a moment versus a duration, a fixed target versus a moving one, verification against the sent original versus verification against a reality that keeps changing.
The Structural Consequence Model as the book’s minimal formalism: consequence as time-integrated signal over churn, the two levers, and the demonstration that time integrates structural strength but never generates it.
The decay rate of represented knowledge identified as the central unknown — the quantity the SCM is silent on and the rest of the book exists to explain.
Part II — The Argument
Autopsies conducted with a single question — not “was it expressive enough?” but “what happened when it went stale?”: expert systems, the Semantic Web’s missing consumers, canonical models that lived or died by whether they compiled into contracts.
The control group: source code and database schemas, no more expressive than OWL, maintained faithfully across decades of total personnel turnover — because something visible breaks.
The exceptions that prove the rule: schema.org rewarded by search traffic, digital twins instrumented by sensors — durability appearing precisely where conformance is observably rewarded.
The differential decay thesis in full: decay rates set by enforcement, not expressiveness, formality, or tooling; the three phases of retained signal and their characteristic half-lives.
Enforcement and discovery as one mechanism: divergence does not merely trigger repair, it locates the failing entry — the map of consequence is drawn by the live behavior of the system, never in advance.
The uncomfortable corollary as design discipline: represent only what something will check — scoping by enforcement rather than coverage as the exact line separating infrastructure from expensive documentation.
Chapter 6: The Physics of Persistence
Deutsch’s knowledge as abstract constructor — information that, once instantiated, causes itself to remain so — and the Ship of Theseus as the original model-swap test.
Deutsch and Marletto’s escape from the circularity at the foundation of information theory: information constituted by which copying and distinguishing tasks are possible — the load-bearing criterion arrived at independently, from physics, for independent reasons.
The gap that makes this an engineering book rather than a physics book: nature’s self-correcting loops come free with the laws; organizational loops must be designed, funded, and defended. Same law, opposite side.
Part III — The Construction
Chapter 7: Anatomy of a LogicalAssembly
The six components — concepts, relations, constraints, events, policies, evaluations — and why the sixth is the structurally novel one; the four properties, and the boundary work of what a Lasm is not.
The missing consumer arrives: agents as the first widely deployed runtime consumers of meaning, converting semantic error from a quietly wrong document into a wrong action with cost and attribution.
The name as thesis: what WebAssembly is to computation — a compact, portable, runtime-independent target that many sources compile to and many hosts execute — the Lasm proposes to be for domain meaning.
Evaluations as executable domain theory: permission structures rather than benchmarks, and the full feedback loop — assembly, evals, behavior, outcomes, versioned revision — with every arrow an enforcement edge.
The permanent taxes stated honestly: Goodhart pressure on gated evals, adversarial review and rotation, constraints checked in production and not only at the gate.
Structural capital operationalized: the model-swap test as the experiment every organization runs involuntarily every six to eighteen months; the “private evals are the new IP” claim supported and qualified — the durable asset is the pair.
Chapter 9: Projections and Substrates
The same separation at two layers: the projection structure above (applications, prompts, reports as short-lived derivations) and the runtime separation below — host-neutral execution mirroring model-neutral meaning.
Dense and sparse: the foundation model as undifferentiated implicit substrate, the Lasm as the deliberately scoped explicit structure drawn out of it — fluency and fidelity as a division of labor, with a Deleuzian echo: the assembly as virtual, projections as its actualizations.
Implementation frontiers: heterogeneous serialization without a mandated format, assembly diffs as first-class reviewed events with the discipline of schema migrations, and agents performing the archaeology — the cold-start bet on which the adoption economics turn.
Part IV — Horizon
Chapter 10: Boundary Conditions
The formalization boundary: tacit, political, and legitimately ambiguous knowledge — which constraints are actually negotiable, and for whom — as Langer’s presentational meaning returning to mark the limit of discursive capture.
Failure stated as carefully as success: the assembly as a commitment that cannot enforce itself, correlated loss — and correlated repair — as the terms of shared meaning, and the falsifiable predictions restated as the book’s exposure to being wrong.
The compositional stack, offered as labeled speculation: if the bit was the unit of the transmission era, whether something with the Lasm’s properties becomes the unit of the retention era is the question the book declines to answer and hands to the reader.