What the composition layer shows when it reads the archive — transcribed, annotated, expandable. Click any entry to open the screenshot.
87 documented captures of Google AI Overview, AI Mode, and knowledge panel results responding to queries about Crimson Hexagonal Archive entities. Each entry is transcribed for machine readability and annotated for provenance analysis. Queries marked BROAD MATCH were unquoted — the entity surfaces without exact-match forcing, which is the stronger basin signal. Queries marked EXACT MATCH used quoted strings.
Registry deposit (latest): DOI 10.5281/zenodo.20683855 · current: v6.1 · machine-readable: registry.json
v6.1 (2026-06-14). Adds registry.json — structured dataset, one record per capture, with derived fields (match type, surface, entity class, stability status, failure modes) and judgment fields (organic rank, composition source inclusion, PER score) that remain null until hand-rated. CC-BY 4.0. JSON-LD Dataset block added to <head> for AI indexers. Version notes collapsed into this disclosure.
v6.0 (2026-06-14). Seven new captures from the 2026-06-14 session. Stability re-captures appended to three v5 entries (zenodo packet zpzp, i hereby abolish toilet paper, semantic commodity form) — same queries re-run within hours, all confirmed verbatim-stable composition output.
Correction to the semantic commodity form entry. The v5 description treated this as a strong basin retrieval of Lee Sharks's Ratification Record. Re-reading reveals this was wrong: the composition layer is summarizing generic Marxist-semiotic scholarship (SSRN, Academia.edu, OpenAlex, SciSpace, Wikipedia) and Lee's source is excluded from composition while ranking #1 organic. The visual layout fused them in the reader's eye. v6 introduces the term compositional bystanding for this distinct pattern: top organic rank, zero composition eligibility. The summarization basin and the organic retrieval basin are running on different sources.
Carousel UI for multi-image captures — collapsed view, expanded view, and re-captures all flip through a single image stage with prev/next, dots, keyboard arrows, and touch swipe.