The process

THE LAW ~14,300 RDAP checks were spent to re-derive (the agency hunt). PROMOTED to the top 2026-08-27 so no session EVER grinds for it again. Full derivation → naming/context/naming-log.md §2.

Four requirements are MUTUALLY EXCLUSIVE — you get any THREE, never four: (1) exact-match .com AVAILABLE · (2) NO aftermarket buy ($0 down) · (3) top-tier / AI-smart / elite SOUND · (4) real dictionary MEANING. Every comparable company broke exactly ONE: PROFOUND = real premium word, domain=tryprofound.com (broke #1) · VERCEL/GROQ/SUNO/FIGMA/VANTA/BREX = coined a non-word (broke #4) · STRIPE/RAMP/LINEAR/NOTION/ANTHROPIC = paid 5-7 figures (broke #2). THERE IS NO FOURTH DOOR.

At step 0 of THE_PROCESS pick WHICH ONE he breaks — that single choice defines the whole hunt: break #1 → the <name><qualifier>.com shape (his Nova→NovaLabs / Profound→tryprofound call: brand word leads, qualifier hq/labs/studio/co/works follows) · break #2 → open the aftermarket (now CHEAP — a good brandable is ~$1-4k, NOT $23k; see AFTERMARKET_PLAYBOOK) · break #3 → a warm real-word compound (the ByteDance / both-parts lane) · break #4 → a CRAFTED coinage (Vercel-grade real-root echo, NEVER mechanical suffix-mash = elvish). Do NOT batch until the door is named — un-named = the 14k-check grind.

TEETH for DO_NOT_REVERT #6 (preached, then violated). HARD STOP: after ~200 gate-0-clearing RDAP checks in one hunt with no survivor he loves, STOP — do NOT open another batch. Output the trilemma verdict ('the free pool for {brief} is empty; the door to break is X') + the 2 real paths, and let HIM choose. A named door + one curated shortlist beats 5,000 more checks. Endless batches ARE the loop he is tired of.

1. DEFINE the buyer + the register FIRST (tool vs agency-house vs consumer; who hears this name and where — invoice line, phone call, cold DM, app store).

2. GENERATE wide (hundreds+; GENERATION_METHODS.volume_rule says 1,000+) inside the gates, IN THE REGISTER, using that register's PRIMARY method — Register C (holdings/agencies/great B2B) → the DISTANT-LEAP BLEND; Register B (tools/products) → the HOUSE_PATTERN.

3. GATE silently (see THE_GATES) — never present a candidate that fails a known gate.

4. VERIFY at the REGISTRY: Bash `whois <domain>` — available ONLY if output contains 'No match for domain'. The instant-domain-search MCP is a CANDIDATE FILTER ONLY — it DNS-checks and FALSE-POSITIVES (two .coms both 'available' by MCP, TAKEN by whois). Never recommend a buy on MCP data alone.

5. BUYER-TEST the survivors: the COLD-READ PROBE first (GATE 0h / GENERATION_METHODS.cold_read_probe — zero-context 'what does this company sell?' across model families; wrong-category mode = dead), then the INVOICE TEST + the phone test out loud. Then answer 'WHY is this still free?' (DO_NOT_REVERT #15) before any pitch.

6. LOCK-AND-BUILD: per his own doctrine — 'when the founder picks the same name three times across alternatives, the cost of a wrong name is smaller than the cost of more deliberation.' A name has ZERO equity pre-revenue; renaming pre-launch costs ~$10. Decide fast, don't loop.

7. REGISTER SAME-DAY (Cloudflare, at-cost). Verified-available ≤8-letter names get sniped within days (observed live: neighboring names registered days before the hunt). Grab 1-2 defensive variants (~$10 each). Quick trademark glance first (Google + EUIPO/WIPO).

TEETH for DO_NOT_REVERT #6 (preached, then violated). HARD STOP: after ~200 gate-0-clearing RDAP checks in one hunt with no survivor he loves, STOP — do NOT open another batch. Output the trilemma verdict ('the free pool for {brief} is empty; the door to break is X') + the 2 real paths, and let HIM choose. A named door + one curated shortlist beats 5,000 more checks. Endless batches ARE the loop he is tired of.