The quadrant nobody mines

Every naming cycle before pipeline v3 mined one of two quadrants, and only two. The first is real words — high neighbourhood density, instantly comprehensible, and therefore catastrophically occupied. In round 13 the desk measured it cold: 24 of 25 real words were already taken on .ai. The free real-word door, whether on .com or .ai, simply does not exist. The second quadrant is the opposite mistake: low-density coinages — free as air but linguistically empty mush, the tidezo/dawnva family that dies the moment you try to tell its story because it has no phonological neighbourhood to resolve against.

The unmined quadrant sits between them. It is lexically empty + phonologically dense: constructed words that are legally blank — no dictionary entry, no collision, no trademark — yet feel like real words because their sound-sequence sits inside a dense neighbourhood of familiar ones. The exemplar is planity, which resolves against plan, sanity, and plain without being any of them. It is a word the ear already half-knows and the register has never seen.

The construction doctrine is the mechanism that reaches that quadrant deliberately rather than by luck. It runs a morpheme matrix (product-real roots crossed with quality tails), pushes candidates through hard filters — a banned-tail screen (no Latin-south tails like ia/va/ea), a slur substring screen, length bounds, and a dictionary-exclusion check so the result is guaranteed lexically empty — and then scores what survives. The score is two terms: character-bigram log-probability plus log(neighbourhood density). The first prizes “how English does this sound”; the second prizes “how many real words cluster around it.” Dense phonotactics plus a populated neighbourhood is precisely the recipe for a coinage that no one has to be taught.

Round 13 is the measured result: 18 of 26 candidates in this quadrant were still free on .ai, against 24 of 25 taken for real words. The domain wall — the assumed hard boundary of the whole market — is gone in this quadrant. The wall was never about domains; it was about where the previous methods kept looking.

The economic incentive to mine this quadrant rather than argue with the aftermarket is independently documented. BrandScout’s 2026-03 study across 4,000 brands found that coined names following natural phonotactics outperform real-word names on both trademarkability (2×) and 48-hour recall (+31%). A constructed-but-natural name is simultaneously easier to own legally and easier to remember than the dictionary word it displaces. The old assumption — that invented names are the compromise you make when the good word is taken — is backwards. In 2026 the constructed name that sounds real is the premium lane, and it is still mostly empty.

None of this is a pipe dream about “creativity.” It is an engineering claim with a score attached: the engine generates, filters, and ranks; the human panel discriminates; and the cold-read probe then checks whether the constructed word lands as a real word in a stranger’s head. The quadrant nobody mines is exactly the one that rewards being mined with a plan.