Six counts the Word Counter actually returned

A word limit needs a counting rule

The same-looking accented text returned three words in composed Unicode and six in decomposed Unicode. A hard return changed a paragraph total without changing the word total. To use either count, you need to know which boundary the counter recognizes—and which one your assignment requires.

The cover shows an observation from 2026-08-17. The ten copyable inputs below were entered on 2026-09-13: six original cases, variants for line breaks and accents, and an emoji. The greeting with a real LF, NFC accents, two-sentence sample, triple hyphen and Chinese sample all matched their August totals. The URL record needs a correction, explained below.

To compare, use language mode auto and writing mode blog. The recorded run used Chromium 148.0.7778.96, UTC, en-US, in a fresh browser context. These are observed counts for those settings, not a linguistic standard.

1. A greeting with one hard return

The first block contains a real line break; the second contains the literal characters \n. Both returned seven words and two sentences, but only the real LF produced two paragraphs.

The character totals differ too. A log may display a line break as an escape sequence, so compare the stored input rather than relying on how the log prints it.

2. The URL case: one line or two?

The 2026-08-17 record reported 3 words, 42 characters, 39 without spaces, 2 sentences and 1 paragraph, but showed a hard return after mother-in-law while describing a single line. That pairing was inconsistent: its paragraph total cannot substantiate the two-line input.

The September variants isolate the difference: replace the space after mother-in-law with an LF and paragraphs rise from one to two. All other listed totals stay the same. The period here is inside the URL, not at the end of a sentence. Treat the sentence field as a heuristic, not a linguistic judgement.

3. The same-looking accents, different totals

NFC stores these accented letters as individual code points. NFD uses letters followed by combining marks for the diaeresis, acute accents and grave accent. The strings are canonically equivalent, yet NFD returned 6 words and 22 UTF-16 units, against NFC’s 3 and 18.

If apparently identical text gets a different count after a paste, compare its Unicode representation before deleting words. Keep the original: normalization changes the stored string. This pair exposes a normalization-sensitive count, not the behavior of every accented word or language.

4. Two paragraphs without a blank line

A single LF separates these sentences; there is no empty line. The counter nevertheless reports two paragraphs. If your assignment defines a paragraph break as a blank line, this total uses a different boundary. Changing the layout just to make the numbers agree would hide that difference.

5. Two words without a space

hello---world has no spaces but returned two words. Counting only space-separated groups would miss this split. Preserve all three hyphens when comparing counters; this result does not tell you how every hyphenated construction is handled.

6. One displayed word is not Chinese segmentation

The entire unspaced sample returned one word. That total does not identify linguistic boundaries inside the text. For a Chinese word limit, use the required segmentation method; for a character limit, first establish the character unit. Neither requirement is answered by calling this string one word.

An emoji makes the character unit visible

The additional September input, 😀, is one Unicode code point stored as two UTF-16 code units. The interface’s 2 characters reflects the latter. A code-point counter would be measuring a different unit.

Neither code points nor UTF-16 units always correspond to the characters a reader perceives. For a character limit, use the destination’s stated unit—or check the exact text there.