A word list, not a word length
Dale–Chall is the odd one out in this collection, and the most interesting. Every other formula infers difficulty from a shape — how long the words are, how many syllables, how long the sentences. Dale–Chall asks a different question: is this word one that a fourth-grade reader actually knows? It answers by checking each word against a list of about 3,000 words established as familiar to fourth graders.
That is why it handles the case the shape-based formulas get wrong. "Grandmother" is eleven letters and three syllables, and every syllable-based formula treats it as hard; it is on the familiar list, so Dale–Chall does not. "Writ" is four letters and one syllable, and no shape-based formula flags it; it is not on the list, so Dale–Chall does.
The score is 0.1579 × (percentage of unfamiliar words) + 0.0496 × average sentence length, with an important discontinuity: if more than 5% of words are unfamiliar, a further 3.6365 is added. That step is in the original formula, and it means a document sitting just either side of 5% can jump most of a grade level for a small change in vocabulary.
How matching works here, and what it forgives
Matching is not literal. Before a word is judged unfamiliar, common inflections are stripped — plural "-s", "-ed", "-ing", "-ly", "-er", "-est" and possessives — so "walked", "walking" and "walks" all match "walk" on the list rather than counting as three unfamiliar words. Anything beginning with a digit is treated as familiar, on the basis that numbers are read rather than looked up.
The words this tool reports as difficult are therefore worth reading rather than trusting blindly. Proper nouns are the largest category of false positives: names of people, places and products are not on a list of general vocabulary and will be flagged every time, even though a reader has no trouble with them in context.
Used well, the list of flagged words is more useful than the score. It is a concrete, per-word edit list, which no other formula here produces.