Construction engine · pipeline v3
Construct names that feel real
Build coined words in the only quadrant left standing — lexically empty, phonologically dense. Scored, not picked on vibes.
Availability is not checked here — a generated name is scored for phonotactics and lexical fit only. Every candidate must still be verified at the registry before purchase. Doctrine: never recommend a buy on an availability API alone.
Why this quadrant exists
Real words
Maximum phonological density, and all taken — dictionary words are exhausted across .com and .ai. Buyability tends toward zero.
Zero-density coinages
Free but mush. A name with no phonological neighbourhood (tidezo, dawnva) has nothing for memory or trust to attach to.
Lexically empty + phonologically dense
Constructed words that feel real — planity reads as plan / sanity / plain — while remaining legally empty. This is where we generate.
The scoring formula
score = char-bigram log-probability + 0.35 · log(neighbourhood density)
Char-bigram log-probability measures phonotactic naturalness — how much the name reads like common English. Neighbourhood density counts real dictionary words within edit distance 2: a name near many real words is remembered and trusted like one.
The evidence
Vitevitch & Donoso, 2012
Higher phonotactic probability raises purchase intent for a candidate word.
Vitevitch–Luce
High neighbourhood-density nonwords are processed and trusted like real words.
BrandScout, 2026
Natural-phonotactics coinages score 2× trademark-ability and +31% 48-hour recall versus real words.
The doctrine entry
Construction engine (pipeline v3)
THE CONSTRUCTION DOCTRINE. Every prior cycle mined one of two quadrants: (1) REAL WORDS (high density, all taken — 24/25 on .ai) or (2) LOW-DENSITY COINAGES (free but mush: tidezo/dawnva = zero phonological neighbourhood). The unmined quadrant: LEXICALLY EMPTY + PHONOLOGICALLY DENSE — constructed words that FEEL like real words (planity ~ plan/sanity/plain; glissando around lissando) while being legally empty. Mechanism: morpheme matrix → hard filters → score = char-bigram log-prob + log(neighbourhood density). Round 13 measured 18/26 FREE on .ai here — the domain wall is gone in this quadrant.