Mechanism
Reference descriptions of the engine's observable behaviour. These entries describe what the system does from the outside without exposing the implementation. Read before the experimental reports to avoid misreading measurement descriptions in later sections.
-
Portfolio ·
A Deterministic English Fluency Layer
A ~50 KB cabinet of rule-based specialists that turn a typed semantic trace into fluent English. No neural network, no training, no GPU. Inflection, lemmatisation with etymological layering, article selection, verb conjugation. 104/104 tests pass. A reference implementation for the separation of semantic navigation and surface rendering.
-
Portfolio ·
Etymological Layering in English Lemmatisation
English is three morphological systems pretending to be one — Germanic inflection, partly-productive Latin derivation, Greek compound roots, French borrowings. A vocabulary-gated collapse rule per layer handles all four correctly; a single uniform rule cannot. The layer distribution of a text also turns out to be a cheap stylometric feature.
-
Portfolio ·
Post-Hoc Morphology Correction for Quantised LLMs
Quantised and distilled language models lose irregular morphology first — 'runned', 'childs', 'mouses' — because the irregulars are carried by a smaller fraction of parameters. A 450-entry irregulars table plus a short repair function catches and corrects these without retraining, without latency cost, and without touching the model. Closed-form failures should not be solved by statistical learners.
-
Portfolio ·
The Conductor
Every transformer's hidden layers build a geometric structure the output layer can't see. A single vector can find it. The model can learn to listen.
-
Portfolio ·
The Geometric Perceptron
A behaviour-first tour of the mechanism that replaces if/elif with continuous decision surfaces.