The WCLM: a words content language model

Briefs 31–34's experiment: an engine in the shape of a transformer where nothing is learned and everything is named. Tokens are content hashes, so the same word tokenises identically in every document. The pipeline is layers of reusable engines — toggle them, drag them between layers, run several side by side in one slot, and each layer reads only the layer before it, so no wire ever jumps. Early engines say this is what we think you said (the dictionary and thesaurus repairs, with their evidence). The senses engine knows a word means different things in different industries — a graph is a network graph here, a chart in a boardroom, a function plot at school — and switching a word's sense shows exactly which of this universe's claims stop applying; singular and plural are read as evidence too, because graph is not graphs. Operators make the little words count: without is not through, and when the prompt negates something this universe asserts, the contradiction is said out loud. Every engine declares its schema — the data types it reads and writes, six types in the whole pipeline — so compatibility is structural: an engine placed where its input type is not yet written is skipped with the reason named, and any engine writing the right type can stand in. It is even fractal: the fractal engine is a full WCLM inside an engine, re-running the winning meaning's own statement one zoom down. The translate engine speaks analogies: pick an audience and the answer is restated in their own concept — graphs of graphs, for somebody from finance, is spreadsheets of spreadsheets (the analogies register) — and every answer now declares its anchoring: a quoted fact, a stated claim, or an authored term. And the query flips: instead of predicting the next word, ask what does this mean — the answer is a concept with its statement, its anchored quote, its blast radius, and a click on any box lights up its full evidence trail. Same prompt, same world, same picture, every time. Training this model means editing its graph inputs (the world, the meaning packs, the senses register), never fitting numbers. Since brief 36 every engine is also a first-class folder: the operators — code, schema, official data with provenance, docs, example vectors, and a workbench each.