The River and Its Rules: Rendering a Corpus's Co-activation and Testing What It Makes

A corpus of authored statements has a shape: which concepts it says together, how often, and which questions run through them. This working paper describes a deterministic render of that shape. Each statement and question is read once into Abstract Meaning Representation, and the render works only from the readings that pass the engine's audit. It draws the concepts as a river grown by space colonization over a force-directed layout, the questions as threads of light bundled along it, and the communities of co-activation as areas coloured through the water. The paper separates what the drawing rules make (how width follows flow, how streams branch) from what the corpus makes (which concepts belong together), and tests the second against the same machinery fed a shuffled corpus. On a private corpus of 309 statements and 170 questions, shown only as shapes, co-activation inside an area is denser than across areas: 12.9 times, averaged area by area, against 6.9–7.8 shuffled. Shuffling widens an activated concept's footprint, and a pretrained embedding model laid over the river mostly differs from it; the design reads each difference as a place to inspect, which this paper does not test. The method is given in full, with its algorithms and constants. The editor that draws the render and the private corpus are not released: the corpus appears only as unlabelled shapes and aggregate numbers. The paper's limits cover projection to two dimensions, the shapes the rules impose, the meeting points' dependence on one setting and one layout, a constrained null, heuristic areas, and concepts read without their relations. An appendix reads the paper's own citations with Gadfly, including a planted false claim as a control; every citation its abstract could not show was checked by hand against the source, and one sentence was corrected. The citation run's records are available from the author on request.