FAQ
Frequently asked questions
The questions below come up every cycle. The answers are the doctrine, not marketing.
Why not just use .com?
Because the free-real-word .com door does not exist, and the doctrine no longer treats .com as absolute anyway. SUPERSEDES the old ".com-only forever" rule (relaxed 2026-08-27): .com fell to ~46% of YC startups and Placek calls .com just an area code. The NAME-quality bar (gate 0) is unchanged — this governs ONLY the TLD once a great name exists, and it is BUYER-MATCHED. The name-quality bar is unchanged — the TLD policy only governs which extension a great name wears, matched to the buyer.
Are AI name generators useless?
Not useless — mis-tasked. LLM-as-generator = LINGUISTIC AVERAGING. Models converge on the statistical centre of "what a good name looks like" and produce consensus mush: Flux / Pulse / Axera. The diagnosis: Method ≈ 70% · brief ≈ 20% · model ≈ 10% of the failure. The fix is that AI is a profiler, never the generator — the engine generates from morphemes and phonesthemes, and the panel discriminates.
Why ≤8 letters?
Because the sweet spot is short and writable. The length gate defaults to ≤8 letters, letters only, one word — softened to 9–10 only when a name is genuinely distinctive with open trademark space. The volume rule cites the same reality: the median modern name is 8 characters, from 6–10.
What is the phone test?
The first filter run on every candidate. Say it once on a phone; the listener must write it correctly with zero spelling help — one word, no homophone-ambiguous letters, no silent letters, no accents. A plain writable name beats a clever unwritable one every time.
Does the score guarantee availability?
No. The score is the machine-checkable half of the doctrine: word-screen gates, length and form, TLD policy. Availability is a separate question answered at the registry, and it ages immediately. Re-check at the registry before acting.
What is the cold-read probe?
Run on every finalist BEFORE the collision sweep — cheaper and kills faster. Protocol: (1) zero-context probes, one question per FRESH session, ≥3 model FAMILIES (not 3 runs of one), ≥5 runs each at default temperature (the SPREAD is the signal); (2) third-person framing always — "a company is called X: what do they sell? what industry?" — NEVER "I am considering naming my company X" (sycophancy caves the test); (3) the shortlist probe: does the name even belong on that register?; (4) score by dominance of the first guess.
Why do some checks say “human judgement”?
Because the doctrine insists they cannot be faked into a number. The first-read law (0h), the buyer invoice test, and the tri-civilisation landing are procedural — they need a real listener, a real statement, a real ear. The desk surfaces them as a checklist rather than auto-scoring them into a false confidence.
What is the free+great+.com lane then?
The distant-leap blend: two real words from far-apart domains resolving into one business — the ByteDance move. Otherwise: buy a two-word brandable, or drop-catch a great expiring .com.
Is the appraisal a trademark clearance?
No. The appraisal clears linguistic and taste gates. Trademark is a separate legal gate (USPTO, TMview, WIPO, a common-law sweep, and native-reader linguistic screens) run as its own knockout path — cheapest first.
What does “lexically empty, phonologically dense” mean?
The construction quadrant. Real words are phonologically dense but all taken; zero-anchor coinages are free but mush. The engine builds words that feel like real words — planity reads as plan/sanity/plain — while remaining legally empty, scored by char-bigram log-probability plus neighbourhood density. That is where the domain wall is gone.
Why is this site not indexed?
Because it is an internal research instrument, not a public product. Every page carries a robots noindex meta and the gateway adds an X-Robots-Tag header. Nothing here is a commercial offer; the desk is a working tool with its sources written down next to it.
What does the distant-leap blend mean in practice?
The #1 method for a free + great + .com name in 2026. The exemplars are BLENDS of two real words, NOT single dictionary words: ByteDance = byte(tech) + dance(art/social); Pinterest = pin + interest; Instagram = instant + telegram; Airbnb = air + bnb. Single real words are all taken; the novel BLEND is free precisely because nobody paired those two words. THE KEY = DISTANT LEAP: the two words must come from FAR-APART domains that resolve into the business (tech↔dance, not trade↔wind). Obvious same-field pairs are taken; the SURPRISE of the leap is what keeps the .com free AND makes it click.