How the difficulty score works, with the arithmetic shown
By Aurangzeb Khan··4 min read
Every word list online sorts into easy, medium and hard, and not one of them says how. The sorting is somebody's impression published as though it were a measurement. This site does the measurement instead, and this piece shows it being done.
What is being measured
Not how obscure a word is. How much its spelling gives away to a standard opening. Those are different properties and conflating them is why most lists feel wrong when you play them.
A guesser opens with the common letters. The question the score answers is: after that opening, how much of the board is filled?
The four measurements
Common cover — what share of the word's letter positions are taken by the eight letters people guess first. High cover means the opening fills the board. Weighted 10, the largest term, because it is the one that decides most rounds.
Vowel share — what proportion of the word is vowels. Vowels are guessed first and nearly always, so a vowel-heavy word surrenders early. Weighted 8.
Rare share — what proportion of the word's distinct letters are rare ones. J, Q, X and Z are reached late if at all. Weighted 6.
Distinct ratio — distinct letters divided by length. A word with repeats gives up more per correct guess, so a low ratio is a kind word. Weighted 3.
Multi-word phrases get a discount of 0.6 per extra word, capped at 2, because the visible spaces hand over structure before anyone guesses.
Three worked examples
RHINOCEROS scores 1.2 — 10 letters, 8 distinct, 80% of its positions covered by the eight commonest letters
DUCK scores 12.5 — 4 letters, 4 distinct, 0% of its positions covered by the eight commonest letters
JAZZ scores 11.8 — 4 letters, 3 distinct, 25% of its positions covered by the eight commonest letters
RHINOCEROS is long and obscure-sounding and scores low, because an opening guess of E or O fills several positions at once and the shape is then enough to name it. DUCK is a word every child knows and scores high: four letters, no repeat, and its only vowel is U — the vowel a guesser reaches for fourth, by which point four wrong guesses are gone.
JAZZ sits high for a different reason: the repeated Z helps a guesser, but only once they have guessed Z at all, and almost nobody does before the board is already lost.
The thresholds
Below 3.6 is easy. From 3.6 to just under 6.4 is medium. 6.4 and above is hard.
The classification is made on the score rounded to one decimal, which is also the number printed in every table. That sounds like a detail and is not: classifying on a finer figure than the page shows produced two words both displaying 3.6 sitting in different tiers, which is correct arithmetic and an obviously broken-looking website.
Negative scores
A few words come out below zero. It means the word gives away more than an average word does — lots of vowels, lots of repetition, almost nothing rare. EERIE is the clearest case: five letters, three of them E.
The scale is not bounded at zero because there was no reason to bound it. Clamping would have hidden the fact that some words are genuinely easier than the formula's neutral point.
What the score cannot see
Whether you have met the word. Whether the category was announced. How good the guesser is. Whether the word has been in the news that week.
All of those matter more at a real table than anything the formula measures. The score is a statement about spelling and nothing else, and that is a smaller claim than the lists it competes with make. It is also the only one that can be checked, which is the trade.
What was considered and left out
Word frequency — how often a word appears in English. It would improve the score for a general audience and ruin it for a category list, where every word is common within its subject and the measure would say nothing useful.
Letter position. Where a letter sits genuinely matters, because guessers reason from endings, but modelling it properly needs more data than this site has and modelling it badly is worse than leaving it out.
Whether the word has a well-known phrase attached. Real, unmeasurable, and the kind of thing a formula should decline to guess at.
Checking it yourself
The weights above are the real constants from the code, printed here by reading them rather than by being copied into prose. The full method, including how rare letters are chosen and how the common-letter set was picked, is on the methodology page.
If you think a weight is wrong, that is a reasonable position and a more useful conversation than arguing about one word's rating. Say which of the four you would change and in which direction.