The original executive-summary site has three flagged violations of the ShurIQ visual grammar: prediction-accuracy framing of the knowledge graph thesis, four "not X, but Y" rhetorical inversions, and an inconsistent wordmark in serif headings. Each version below applies one fix in isolation, so the effect of each rule can be read on its own.
Current state of projects/shur/executive-summary/index.html. No edits applied. The reference point all three fixes diverge from.
Reframes the lead numbers from accuracy points to graph depth. Tab 3 shifts from "Autoresearch Pipeline" to "Nightly Intelligence Operations" with three concurrent products: detection, defense, calibration. Calibration (the accuracy-tuning track) is positioned as one signal test among several, not the justification for the graph.
Four targeted rewrites. The "not X, but Y" pattern and definitional negation ("isn't whether... it's how many...") are the most obvious AI-generated-text tells; readers register them as machine output the moment they appear.
.brand wrapped"SHUR IQ" (uppercase, spaced) reads to a serif heading as a separate two-word phrase. Canonical wordmark is "ShurIQ" rendered in IBM Plex Sans even when surrounded by serif type.
.brand (sans) inside any serif heading. Canonical wordmark is ShurIQ, never "SHUR IQ" with a space. Don't hand-roll Playfair stacks. Use IBM Plex consistently — serif for body and headings, sans (via .brand) for the wordmark.