String Distance

Features

• Edit distance, with a character-by-character alignment and a list of what changed (which substitutions, insertions and deletions)
• Damerau-Levenshtein: an adjacent swap is one step, so `teh` → `the` is 1 rather than 2
• Hamming distance: ⚠️ when the lengths differ it says so instead of inventing a number, because the measure is not defined there
• Jaro and Jaro-Winkler similarity (0–1), prefix capped at 4 characters and the bonus applied only above the original paper’s 0.7 threshold
• Longest common subsequence: gives the length and the subsequence itself
• ⚠️ The similarity row carries a note that it is a convention, so it is not compared against Jaro-Winkler
• CJK characters are padded to two columns so the alignment does not drift
• ⚠️ A string over 5,000 characters is reported rather than computed — the matrix would hang the tab, and a frozen result helps nobody
• Case sensitivity is optional, and normalising only affects the comparison, never the strings shown back

How to use

1. Put A on the first line and B on the second
2. Pick the algorithm — they answer different questions
3. The edit-distance path gives a character alignment you can work from

FAQ

Why five algorithms instead of one similarity number?

Because they answer different questions, and offering one would mislead. `teh` and `the` differ by 2 under edit distance and by 1 under Damerau — which is right depends on whether you are counting keystrokes or changed letters. Hamming allows substitutions only and requires equal lengths, which is the DNA and bit-error case. Jaro-Winkler returns a similarity rather than a distance and leans towards “the same name if it starts the same”. Pick the wrong one and the number still looks like a number, but it no longer answers your question.

Why does the similarity need a note saying it is a convention?

Because a distance has an exact definition (how many edits) while a similarity does not — there are several ways to fold a distance into 0–1. This one is `1 − distance ÷ longer length`. Jaro-Winkler defines its own, so the two legitimately disagree on the same input. Without saying so, people compare the two numbers and conclude one of them is broken.

Why give no number for Hamming when the lengths differ?

Because it is not defined. Hamming distance is “at how many positions do two equal-length sequences differ”, and with unequal lengths that question does not exist. A number could be manufactured (truncate to the shorter, say), but it would answer nothing while looking authoritative. So the tool says so instead.

Why refuse very long inputs?

Because these algorithms all build an n × m matrix. Twenty thousand characters on each side is four hundred million cells — it exhausts memory and then hangs the tab. String comparison is for short text anyway: typos, names, single-line identifiers. Pasting two long documents here usually means the wrong tool, so past 5,000 characters the tool says so rather than freezing.

Why two lines for the input?

Because one text box holds one string. The second could live in an option field, but comparing strings is a process of trying things, and having to leave the input area on every change is painful. The cost is that a string cannot contain a newline — a deliberate trade: the algorithms handle multi-line strings perfectly well, but “character number N” stops being pointable in the interface.