How we score a name
The name score is not a black box. It is one formula with six terms, and every weight, threshold, and word list behind it is published here — including the places where the model stops being useful.
The score in one paragraph
The number at the top of a result is DTRI, the Domain Typing Risk Index, on a scale from 0 to 100. A 100 means nothing about the name is likely to be mistyped; a score in the fifties means you should budget for explaining the spelling on every call. It is clamped at zero, so a name can never score below zero no matter how many separate things are wrong with it.
DTRI measures one thing: how reliably a person who has heard the name, or half-remembers it, converts it back into the right characters. It deliberately does not measure whether the name is good.
The formula
Six terms, each weighted, each subtracted from a perfect 100:
dtri = 100 − (p × 15) − (s × 10) − (d × 5) − (r × 15) − (l × 10) − (t × 10)
Each term is a risk unit between 0 and 4, not a raw count of problems. Every detector reports a 0–100 risk, and 25 points of risk is one unit — so one clear problem costs one unit, and a term is capped at 4 units so a single catastrophic dimension cannot swallow the whole score. A term can therefore cost at most 4 × its weight in points.
The six terms
p Sound-to-spelling
weight 15 · costs up to 60 points
Sounds with more than one common spelling. The “ph” in “phone” could be “f”; the “c” in “city” could be “s”. The scan finds every grapheme in the name that a listener could reasonably write a different way.
Detector: P2G ambiguity
s Silent letters & schwas
weight 10 · costs up to 40 points
Letters nobody says out loud, and vowels swallowed into a schwa. The “k” in “knight” never reaches the page because it never reaches the ear.
Detector: Silent letters
d Keyboard slips
weight 5 · costs up to 20 points
Letters close enough together on a phone keyboard that one thumb slip changes the name, plus the traits a keyboard cannot express at all: hyphens and digits.
Detector: QWERTY adjacency
r Repeated letters
weight 15 · costs up to 60 points
Doubles and triples that get collapsed when someone types from memory. Each pair is weighted by how often English actually doubles that letter, so “tt” costs more than “zz”.
Detector: Repeated-letter risk
l Phonetic leak
weight 10 · costs up to 40 points
A plausible respelling of the name that is already a real word, so autocorrect, search suggestions, and voice input reach it first. This is the term that turns a misspelling into someone else's traffic.
Detector: Phonetic variants
t Extension
weight 10 · costs up to 40 points
The cost of the part of the address nobody can infer. “.com” is supplied by the listener; every other ending has to be said out loud (“dot io”), spelled when they hesitate, and typed - and the listener who leaves it off reaches whoever owns the “.com”.
Detector: TLD risk
A note on the formula
The published four-term version of this index counts sound-to-spelling ambiguity, silent letters, keyboard adjacency, and repeated letters. This implementation adds two. l is the phonetic-leak term, weighted equally with the first term: a respelling that is already a real, more common word is the single most common way a name loses a searcher — “tyme” becoming “time”, “kat” becoming “cat” — so it is scored alongside the ambiguity it comes from rather than reported as a footnote. t is the extension term: the part of the address a listener cannot infer, where “.com” is supplied for you and every other ending has to be said out loud. Both are deliberate deviations, and they mean our numbers are not comparable with published DTRI tables.
The grade bands
The label shown next to the number comes from these bands. The boundaries are a judgement call about how much spelling support a name needs, not a natural law:
- Excellent
- 85+
- Say it once and people can find it.
- Strong
- 78+
- A short, deliberate spelling note covers the rough edges.
- Needs work
- 58+
- Expect to spell this name out loud, and to buy the typo domain.
- Risky
- 0+
- Most of your audience will end up somewhere else.
What is deliberately not in the score
A score is defined as much by what it excludes. These are reported alongside the number, or not measured at all, because folding them in would make it answer a different question:
- Distinctiveness. How crowded the spelling neighbourhood is, reported as its own score. A name can be easy to spell and forgettable, or hard to spell and unforgettable.
- Meaning and taste. The score cannot tell you whether a name suits your market, or whether it is embarrassing in a language you do not read.
- Trademark and availability. Whether you can legally use the name, and whether the domain is free, are separate questions with separate tools. Scoring a name says nothing about either.
- Value. This tool scores a name you already have. To see what a listed domain is worth, see how we value domains.
Where the data comes from
Frequency list
The ~10,000 most frequent American English words, in frequency order (the public-domain google-10000-english-usa list), plus hand-added words that matter to naming but sit just outside it, such as “phonetic”. It backs both the “is this a common word?” test and the neighbourhood scan.
Word list
The same ~10,000-word frequency list (USA spellings), also used for category classification on this site. It answers “is this spelling a real word at all?”, which is a different question from “would a keyboard suggest it?”.
Keyboard geometry
A QWERTY layout measured in half-key units, so the distance between any two letters is a number rather than a guess. Adjacent pairs are the ones thumbs slide across.
Phonetics
A simplified Metaphone code plus Soundex and Refined Soundex, used to generate the respellings a listener or a voice keyboard might produce — keeping only the ones whose rewritten letters form a real word — and then to check those respellings against the two word lists above.
Extension table
A short, hand-written list of domain endings and what each costs to communicate: “.com” is what a listener assumes, “.org” and “.net” are understood but still have to be said, “.io” and “.ai” are familiar to a startup audience and ambiguous to everyone else, and “.co” is one character from “.com”. Everything else is treated as a suffix that has to be spelled out loud. It is a judgement call about how often a stranger fills in “.com”, not a measurement.
Edit distance
Damerau–Levenshtein distance (optimal string alignment) against the frequency list, which produces the OLD20-style neighbourhood figure used by the distinctiveness score.
Where the model stops being useful
- It is a model, not a listener study. The detectors encode how English spelling works; they are not a measurement of how any real group of people behaved. Treat a score as a well-informed prediction.
- The neighbourhood is approximate. Published OLD20 figures are measured against a large psycholinguistic corpus. Ours is computed against the ~10,000 most frequent words, restricted to candidates within two characters of the name's length. It is consistent and explainable, but it is not comparable with a published OLD20 table.
- It knows one language and one keyboard. The word lists, silent-letter patterns, and key map are American English and QWERTY. A name that is hard to spell in English may be trivial in Spanish, and a phone keyboard in another locale has different neighbours.
- It guesses at stress. Whether a vowel reduces to a schwa depends on the stress it carries, and spelling does not reveal that: the “-er” of “server” is a schwa, the “-er” of “refer” is not. The detector reports a possible reduction, only away from the first syllable, and it cannot tell you which of the two you are holding.
- It does not know your audience. If every customer hears your name in a podcast, radio-test failure matters enormously. If they arrive from a link, it barely matters at all. The score cannot weigh that for you.
Frequently asked questions
Does the score use AI to guess how people will spell my name?
No. Every part of the score is a deterministic rule that you can read on this page: a list of spellings for each sound, a set of silent-letter patterns, a QWERTY key map, a table of doubled-letter frequency, and the phonetic code of the name. The same name always produces the same score.
Can I reproduce the score myself?
Yes. The formula and every weight are printed above, and the tool shows the breakdown it used: each detected problem, its severity, and the points it cost. If a score surprises you, open the full breakdown and the arithmetic is visible line by line.
Why did my name score well on distinctiveness but badly overall?
Because distinctiveness is not part of the score. Distinctiveness measures how crowded the spelling neighbourhood is - how close the name sits to other English words. That is a brandability signal, not a mistyping signal, so it is reported next to the score rather than folded into it.
Is a low score a reason not to buy a name?
Not on its own. A low score predicts that people will misspell the name, which is a cost you can plan for: buy the misspelling, publish the spelling, and budget a little paid search. It says nothing about whether the name is distinctive, available, or legally clear.
Now score a name and read the breakdown
Every detected problem is listed with the points it cost, so you can check the arithmetic against the formula above.