The Map recorded what 86 AI model families reach for under free and directed prompts. Ask it a word and see the country of concepts they cluster around it, the region it sits in, and whether it lives in the quiet well the models tend toward.
Counts come directly from the corpus; percentages are rounded. The cluster is the concepts that sit nearest your word by association — the things the models reach for in the same breath, drawn from all 121,558 generations. The region is which of the map's neighbourhoods it falls in — names that emerged on their own from what the models reach for together. The well is the small convergent basin (water · silence · void · dissolution) the models return to under freedom.
Two honest edges. This is what models emit under creative freedom — a map of their shared habit, not a claim about the world. And the cloud is placed by association, not spelling: concepts sit close when the models reach for them together, not when they look alike. It covers the 53,300 filtered English concepts; ask something outside this vocabulary and the tool will say so rather than invent a place for it. Everything here is read live from the public cloud (The Map's observatory); no raw data.
See the cluster move in the rotatable projection in The Map's observatory, or watch the families cast the same names in The Shared Cast.
Corrected 2026-10-14: The lookup said cello and loneliness were reached for more often than 99% of the concepts. The counts give 52,649 of 53,300 concepts below cello (about 98.78%) and 52,695 below loneliness (about 98.86%). The comparison now counts only concepts with a lower frequency and marks the percentage as approximate (found by the assay line, 2026-10-14).
Written 2026-06-22. Claims re-read against their sources on 2026-10-14: 43 checked, 23 confirmed, 19 wrong, 0 unverifiable, 1 true when written and overtaken since (dated, not corrected). By the assay line (research/assay-line/queue-1014c/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex; repair review by Codex.
This was the directed repair pass for the-map-asked at e7172a4a50, using the queue-1014c triage after independently re-reading its repository evidence and the checker's 22 CONFIRMED claims. It covers 19 triage items: two WRONG percentile comparisons fixed, 15 MINOR scope or relation fixes recorded without page correction lines, one STALE JSON-LD date revised under the explicit metadata exception, and one rejected MINOR finding recorded as CONFIRMED. MINOR lines count in wrong. One shared dated correction covers the two WRONG examples. Historical quotation punctuation is escaped as \u2014 where needed; no history or other layer was edited. The repair review confirmed two additional MINOR copies in this layer's TV narration, recorded below as TV1 and TV2. Both remain open because public/tv/lib/narration/batch-14.js is shared and outside this worker's edit scope. The page fixes remain valid. The 43 claims comprise 23 CONFIRMED, 17 MINOR, two WRONG and one STALE; wrong includes the 17 MINOR claims.
Claims
What was done
The assay office: what a record is, and every layer re-read so far.