Methodology
Every score is a quality triage signal, not a magic appraisal. This is the current v1 scoring contract: base signals first, then hard caps for safety, spam, trademarks, and false-premium patterns.
DQS v1 starts with readable signals, then applies governors. It isn’t a black-box ML model. We hand-pick signals that map to how domainers actually value names, then cap outputs that would otherwise be misleading: random pronounceable strings, keyword-plus-digits spam, hyphen variants, and reputation-risk domains.
Authoritative scoring runs in our daily ingest (Python, services/scorer/). The /appraise on-demand scorer uses a TypeScript port for quick checks. Corpus scoring is the source of truth; the lite scorer is kept close and guarded with edge-case tests, but the daily pipeline owns ranked inventory.
weights sum to 100
How short the SLD is, on a curve.
3–5 chars: full value (1.0). 6 chars: 0.92. 7: 0.82. 8: 0.70. 9: 0.55. 10: 0.40. 11: 0.25. 12: 0.12. 13+: 0.04. Front-loads value in the 3–5 char range because that's where the market treats length as categorical.
Premium level of the TLD the drop lands on.
.com: 1.00. .io: 0.82. .ai: 0.75. .co: 0.60. .app: 0.58. .dev: 0.52. .net: 0.50. .org: 0.42. .xyz: 0.25. .info: 0.15. Default for unknown TLDs: 0.20. We add TLDs to this table only when our comp data supports the specific multiplier.
Can a human say it out loud without spelling it?
Vowel-ratio peak at 40% vowels (natural English phonotactics). 3-consonant clusters fine; 4+ penalized. Catches gibberish like 'krxtfly' while rewarding real-word-shaped SLDs.
Real English word in the SLD.
Uses log-scaled word frequency from a large English corpus. Stems for plurals (-s) and gerunds (-ing). 'shop', 'rails', 'summer' score high. The on-demand /appraise scorer uses a 600-word common-English set as a lite fallback (known ±5 point drift).
Contains a term from a commercial-category vocabulary.
AI, crypto, finance, SaaS, ecommerce, health, real-estate term libraries. Low weight on its own — just a breadcrumb that the domain is adjacent to a buyer cohort.
Exact-match commercial phrase composed of category + modifier.
Fires when the SLD is 'category + intent-word' (aitools, cryptohub, medfinance) or 'category + category'. This is the heaviest positive signal because exact-match commercial-intent domains are what serious buyers actually pay for.
Prior web-archive snapshots — built-in brand context.
0 snaps: 0. 1: 0.25. 10: 0.45. 50: 0.65. 200: 0.85. 500+: 1.00 (linear between points). A domain with 15 years of history carries context a blank-slate SLD doesn't — ships with link equity + brand familiarity.
Short + pronounceable + clean structure = a brandable candidate, not automatically a premium name.
The v1 scorer rewards clean phonetics, but v1.8 adds a premium-band governor: a pronounceable invented name cannot sit in the 70+ band unless it also has clear commercial meaning, exceptional brand shape, short-domain scarcity, or strong TLD/category fit.
subtract up to 255
Proportion of digit characters in SLD.
Linear penalty to 30% digit density. Digits reduce resale value significantly unless they're semantically meaningful (3d, 24, 360).
Hyphens anywhere in the SLD.
1 hyphen → ~0.5 penalty. 2 → ~0.95. 3+ → 1.0 hard penalty. Rare exception: compound words some markets tolerate ('co-op').
SLD over 12 characters.
Zero below 12. Linear ramp to 1.0 at 20 chars. At that point the length-curve positive is already at its 0.04 floor; the penalty stacks it.
Digits in low-trust patterns, spam phrases, and weak keyword stuffing.
Keyword + random digits is capped aggressively: game888hub.com, cashbonus777.com, loan247fast.com, and similar patterns should not escape into the review band. Three or more consecutive digits anywhere in a spam-sensitive phrase triggers a hard ceiling.
Ends in ~bot or ~ify on a recognizable category stem.
'cryptoify', 'gptbot', 'aibot' fire; 'spotify' doesn't (stem isn't a category term). Saturated suffixes are brand-risky even if keyword-rich.
Unsafe language, slurs, sexual-violence terms, violent-crime wording, or obvious brand collisions override positives.
This cap sits above the normal weighted score. Domains like openaiwallet.com, teslapayments.com, offensive slurs, or violent terms do not get rescued by .com, length, or pronounceability. They are suppressed for public quality and advertiser/payment safety.
DQS 70+ requires a real premium qualifier.
A name needs at least one strong qualifier to remain 70+: exact commercial phrase, strong dictionary/category fit, scarce short-domain pattern, exceptional brandable quality, or external proof. Weak med/pay/hub/buy token matches and generic invented names are capped into the 30–60 range.
# Base score, then caps.
positive = sum(signalval × positiveweight)
penalty = sum(penval × penaltyweight)
raw_score = round(max(0, min(100, positive − penalty)))
score = min(raw_score, applicable_caps...)
# Hard rule: all-digit SLDs are handled as a special liquidity class, not a normal brand-quality win.
# Hard rule: DQS 70+ requires a premium qualifier. Pronounceable alone is not enough.
# Hard rule: safety, trademark, and spam caps override every positive signal.
Every row in the drops feed ships with a sentence like “brand-ready shape with commercial finance intent and short”. We refuse to ship a bare number. If you can’t look at the score and explain it to a partner or client, the score is doing work you can’t defend. That’s a worse position than no score at all.
The explanation is generated from actual scoring signals and constrained by lifecycle state. If a score came from an older model or a weak signal, dossier wording stays conservative: “worth reviewing” instead of pretending every high-ish number is a premium asset.
Comparison
Both return a single-number appraisal from an ML model. Both are opaque. Neither ships a per-signal breakdown or an explanation, and the industry-wide joke about them is that any serious domainer prices their opinion at ~10× the quoted number. They exist because nothing better does. DQS is a discovery filter for “is this drop worth the next 15 minutes of my research” — not a valuation. If you want a number to haggle against, treat DQS as an input to your own judgment, not a replacement for it.