Corpus structure · CAV v1.3
Every tracked theory is coded as one four-slot sentence — an actor, doing an act, to an element, for an intent. The four vocabularies multiply out to 16,335 possible sentences; the published corpus says 278 of them. Theories that share a sentence — or come within one slot of it — form the islands below.
Two theories are linked when they share the same element plus at least two of the other three slots. Every link therefore carries the element with it, which makes each island element-pure: one subject per cluster. Drag to pan, scroll to zoom, click a node for its kin and a link to its page.
The eight largest clusters carry their own colour on the map and are direct-labelled on it. Everything smaller shares one muted slate; theories with no archetype sibling are drawn hollow. Each cluster is named by the element it is pure in, plus its most common act.
Every four-slot sentence the published corpus actually says, with how many theories carry it and the loudest of them by name.
Hand-written patterns with one slot left open — a coarser grouping than a full tuple, used to name a neighbourhood rather than a point.
What is coded. Each theory gets one tuple in scto_theory_archetype under
codebook CAV v1.3: element (what the claim is about), actor type, act, intent. A theory whose
claim has no actor doing anything is coded ASSERTION rather than PLOT;
the similarity rule runs on plots only, so assertions appear here unlinked by construction.
How the map is built. Links come from v_scto_shares_archetype: same element
plus at least two of the other three slots. Islands are connected components; positions are a
Fruchterman–Reingold layout run per island and packed on a spiral, so the picture is stable
across loads rather than re-randomised.
Node size is believer-side volume — advocacy claim links, the same count the rest of the site calls atoms. It is loudness, not importance.
The halo is dehumanisation heat: the mean dehum score over a theory's
believer-bin atoms, out-group framed as sub-human rather than merely hostile. It is a
sparse signal — 148 of 394 theories here carry any of it, and the
corpus median is zero. No halo means the framing was not found, not that the theory is benign.
The rim is identity-fusion heat: the mean f_fusion score over the atoms the
FTM battery actually scored. FTM runs behind a recall pre-gate, so most atoms never reach it:
183 theories here have any FTM-scored atoms at all and 57 carry positive fusion.
No rim means not measured, not cold.
Coding provenance. One pass on 2026-08-13 — a constrained-enum model pass with a hand-adjudicated remainder. One coder, so the available reliability figure is test–retest stability, not inter-annotator agreement. 152 coded theories are withheld from this page by the publish gate (proto, merged, or held in integrity review) and are absent rather than unlinked.
Narrative-level triage structure — not a prediction, and not a claim about any individual.