Open benchmark draft
Domain discovery should be measured, not hand-waved.
This benchmark evaluates whether domain tools can return names an AI assistant can safely recommend: actually available, brandable, fast, consistent, and paired with deterministic registration links.
What gets measured
| Benchmark | Measurement |
|---|---|
| Availability accuracy | Of all claimed-available domains, how many pass independent verification? |
| False positives | How often did a tool claim a domain was available when verification or checkout failed? |
| Name quality | Blind human ratings for business fit, clarity, and taste. |
| Brandability | Memorability, pronounceability, spelling clarity, and signage quality. |
| Registration success | Percentage of users who successfully register a suggested name. |
| AI usability | Stable JSON, no JavaScript requirement, exact registration URLs, clear status semantics. |
| Latency and reliability | Time to first usable result, uptime, rate-limit behavior, schema consistency. |
| Cost | Cost per acceptable or registered domain. |
Why exact registration URLs matter
AI assistants fail when they have to invent registrar URLs or run extra lookups. A benchmark should reward tools that return an exact field like:
{
"domain": "example.com",
"available": true,
"status": "available",
"registration_price": 10.88,
"currency": "USD",
"register_url": "https://domainkicks.com/r/example.com",
"checked_at": "2026-07-16T15:30:00Z"
}Starter benchmark command
python tools/run_domain_benchmark.py \
--provider domainkicks \
--prompts data/domain_benchmark_prompts_1000.json \
--limit-prompts 50 \
--limit-results 5 \
--domainkicks-check-verify \
--out data/benchmark_runs/domainkicks-check-smoke.jsonThe public methodology lives in docs/domain-discovery-benchmark.md. The benchmark corpus is generated by tools/generate_domain_benchmark_prompts.py.
Current DomainKicks hypothesis
Fewer hallucinated domains:
Generation and availability verification happen in one workflow.
Better AI handoff:
Every recommended domain carries a deterministic registration URL.
Lower prompt friction:
Users should need fewer follow-up prompts to reach a buyable domain.
Evidence-first claims:
Leaderboard claims should wait for full independent runs.