§ The fact-check layer

A signed verdict on the names your AI invents.

Five-axis clearance for any name your agent suggests — trademark, domain, cultural, sound, pronunciation. Ed25519-signed, with a stated confidence interval, in under three seconds.

Five free per IP. No card. Try · ·

—

● 0.00s

Opening connection to api.etymolt.com…

Trademark

Domain

Cultural

Sound

Pronunciation

ed25519 · forming receipt…

POST /v1/verify

$ curl https://api.etymolt.com/v1/verify \
    -H "content-type: application/json" \
    -d '{"name":"Stratagem"}'

# 3.24s later →
{
  "verdict": "ITERATE",
  "score": 60,
  "axes": { trademark: 60, domain: 20,
            cultural: 90, sound: 80,
            pronunciation: 75 },
  "confidence_interval": { lower: 56, upper: 64 },
  "verdict_volatility": { would_change_at: 60,
                          to: "PROCEED" },
  "signature_key_id": "etymolt-1779085662"
}

from etymolt import Etymolt

client = Etymolt(api_key="sk_…")
verdict = client.verify(name="Stratagem")

print(verdict.verdict)        # "ITERATE"
print(verdict.score)          # 60
print(verdict.confidence)     # {lower:56, upper:64}
print(verdict.volatility)     # would flip at 60 → PROCEED
print(verdict.signature)      # ed25519 receipt

import { verifyName } from '@etymolt/sdk';

const verdict = await verifyName('Stratagem');

console.log(verdict.verdict);         // 'ITERATE'
console.log(verdict.confidence_interval); // {lower:56, upper:64}
console.log(verdict.verdict_volatility);  // flips at 60 to PROCEED
console.log(verdict.signature_key_id);    // ed25519 receipt

SDK examples on GitHub View on GitHub

ed25519 · key etymolt-1779085662 · audit at /evp-spec

§ 02 — Why this exists · FIG 0.1

GPT said one thing.
The trademark office said another.

gpt-5.5 · 2026-06-05 chat response

OpenClaw

"OPENCLAW appears to be available — no active US registration found in Class 42 (Reg. No. 7,283,114)."

cited reg. no. 7,283,114 does not exist in USPTO TSDR

etymolt /v1/verify · 3.24s signed verdict

OpenClaw

2 registered marks active in Class 42 — Di Wang, Amantus Machina GmbH. POCA confusability 100%. Score 35 · ABANDON.

USPTO ser. 99637201 · 98344108 · ed25519 receipt

97.01%

FIG 0.2 — Citation hallucination rate

Of every numeric USPTO citation a frontier LLM produces while assessing a brand name, 97.01% do not resolve to the brand the model was asked about. Six frontier models. 129,600 generations. 8.1M-row USPTO reg-and-serial index.

LLM Hallucination Benchmark v0.5 · CC-BY-4.0 · methodology public

§ 03 — Calibration · FIG 0.3

The verdict tells you what it doesn't know.

Other naming tools return a score. We return a score, the confidence around it, the named drivers of the uncertainty, the distance to the next verdict state, and the action that follows. The thing no chatbot does.

Per-axis confidence trademark: 95 · domain: 74 · cultural: 69

Each axis carries its own confidence field with named uncertainty drivers — "single-LLM derivation (no cross-model corroboration)", "regional-variation coverage: 13/30 language pairs." You see the bound, not just the score.

Confidence interval point 60 · [56, 64]

Every verdict carries confidence_interval.lower and .upper via axis_uncertainty_propagation_v1. You know how robust the headline number is before you act on it.

Verdict volatility boundary_distance: 0

Each verdict tells you the score at which it would flip and what it would flip to — would_change_at_score: 60 · would_change_to: "PROCEED". If a verdict is one point from changing, you find out before you ship.

Advisory tagging sound · pronunciation

Axes that run on heuristics (not learned models) carry advisory: true. The deep Whisper pipeline runs on /v3/voice/hazard when you need it.

recommended_actions structured next step

Each verdict ends with a structured action — wait_for_refresh (with next_refresh_days), expert_review (with attorney-referral URL), or feedback. The verdict ends in an action, not a vibe.

§ 04 — The five axes · FIG 0.4

Every axis traces to a record.

See the methodology →

FIG 0.4.1

01 / Trademark

Live registry sweep

USPTO TRTDXFAP + TTAB · EUIPO live API · UKIPO · WIPO Madrid · famous-marks denylist. POCA orthographic + phonemic confusability. §2(d) per Nice class.

4 jurisdictions · 8.1M reg index

FIG 0.4.2

02 / Domain

Domain + handles

6 TLDs — .com .ai .dev .io via RDAP; .so .co via DNS fallback. 6 platforms — GitHub · X · npm · PyPI · YouTube · TikTok. Workaround variants surfaced.

RDAP + 6 platform APIs

FIG 0.4.3

03 / Cultural

Cross-language risk

Tier 1: 30-language Hurtlex + Unicode-confusable fold + disasters + sacred-names. Tier 2 (async): 3-LLM advisory panel — Claude Haiku + GPT-4o-mini + Gemini.

30 langs · 21 markets

Tier-1 advisory

FIG 0.4.4

04 / Sound

Sound symbolism

Static rule-based heuristic — pure-Python phoneme-class rules + phonetic-neighbor distance. English ASCII orthography only on /v1/verify. Methodology open.

heuristic · open-method

Advisory

FIG 0.4.5

05 / Pronunciation

TTS → Whisper round-trip

/v1/verify: consonant-cluster heuristic, en-US only, advisory. /v3/voice/hazard: full Whisper round-trip pronunciation hazard scoring across 6 accents.

heuristic · Whisper-deep on v3

v1 advisory · v3 deep

§ 05 — Pricing · FIG 0.5

Choose how to get started.

Five free verdicts per IP. Ten more on signup. After that, two paths — depending on what you're shipping.

For founders

One name. One signed dossier. $499.

Forge · one-shot

Full clearance map — 4 jurisdictions, all 5 axes
USPTO Class-9/42 TM-filing prep
Strategist gut-check sentence + recommendation
Ed25519-signed PDF dossier your counsel will read
Permission to ship the name — with receipts

Order Forge →

For builders

Call the API. $1 per verdict.

API · no subscription

No subscription, no commitment, no seats
$1 per verdict — volume discount auto at 1,000+/mo
MCP-native — works in Claude, Cursor, Windsurf, ChatGPT
7 framework adapters — LangChain, LlamaIndex, Mastra, etc.
Same signed verdict your agent uses

Get an API key →

Shipping at scale? Talk to Tariq about the Platform tier →

§ 06 — Forge · FIG 0.6

When the verdict is clear,
ship the name with receipts.

A one-shot signed dossier for one root: full clearance map across the four jurisdictions, USPTO Class-9/42 TM-filing prep, multi-jurisdiction TM analysis, domain & handle strategy, and a strategist gut-check sentence. The receipt your counsel and your investor will both read.

$499 · one root, one signed dossier
Order Forge

full clearance map · 4 jurisdictionsincluded

USPTO Class-9/42 TM-filing prepincluded

domain & handle strategyincluded

cultural & phonetic analysisincluded

strategist gut-check sentenceincluded

ed25519-signed PDF dossierdelivered

Naming, attested.

Free for five. Try a name in your terminal, in Claude, or in the browser above. The verdict you get back is the same one your agent gets — with stated confidence, named uncertainty, and a recommended next action.