// admin tools · 04
Password strength
See how an attacker would actually crack a password: zxcvbn score, guess count, crack times online and offline, the dictionary words and keyboard patterns it found, and how to fix it.
// runs locally — your password never leaves this browser
Never stored, never logged, never sent. Analysis runs in this tab; the field is wiped when you leave.
zxcvbn dictionary loads after your first keystroke.
time to crack
how zxcvbn sees it
dictionary / l33tkeyboardrepeatsequencedatepatternrandom
| pattern | token | detail | guesses |
|---|
feedback
When ticked, the password is hashed with SHA-1 here, and only the first 5 hex characters of that hash are sent to api.pwnedpasswords.com (k-anonymity). The API returns every hash suffix sharing that prefix, padded with decoys, and the match is done in this tab. Like any request, it reveals your IP address and browser to the API.
How it works
What it does
This checker estimates how many guesses an attacker would need to find a password and turns that into crack times for realistic attack scenarios. It uses zxcvbn 4.4.2, Dropbox's open-source estimator, which looks for the shortcuts real cracking tools take: common passwords, English words, names, keyboard walks, repeats, sequences, dates and l33t substitutions. The match sequence shows exactly which part of your password fell to which pattern.
Your password never leaves the browser. The dictionary (about 800 KB) is a static file loaded from this site after your first keystroke. Until it arrives you get a basic charset estimate. Nothing is stored, logged or put in the URL.
How the estimate works
zxcvbn finds every pattern match in the password, then picks the cheapest combination of matches that covers the whole string. Each match gets a guess count. A dictionary word scores its rank in a frequency list, plus extra for capitalisation and l33t. Anything unmatched is scored as brute force. The product of those counts is the guesses figure, shown as a power of ten.
The naive charset entropy line is the textbook formula: length × log2(character pool). It treats Password1! as if every character were random, which is why it is always optimistic. The gap between the two numbers measures how predictable the password is.
What the crack times mean
- online, throttled
- 100 guesses an hour against a live login with rate limiting or lockout.
- online, unthrottled
- 10 guesses a second against a service with no rate limiting.
- offline, slow hash
- 10,000 guesses a second against stolen hashes protected by bcrypt, scrypt or Argon2.
- offline, fast hash
- 10 billion guesses a second against unsalted MD5, SHA-1 or NTLM on GPU rigs.
The score runs from 0 to 4: below 103, 106, 108 and 1010 guesses, then above. These are order-of-magnitude estimates, not guarantees.
The optional breach check
Tick Have I Been Pwned and the page computes the password's SHA-1 locally, then sends only the first five hex characters to api.pwnedpasswords.com/range/. The response lists every breached hash with that prefix, plus padding, and the comparison happens here. The service never sees your password or its full hash.
FAQ
Is it safe to type my real password here?
The analysis is local and the page makes no network requests unless you opt in to the breach check. Even so, the safest habit is to test a password of the same shape rather than the real one.
Why does a long passphrase beat a short complex password?
Guess counts multiply. Four random common words from a large list give far more combinations than eight characters of l33t-speak built on a dictionary word, which zxcvbn and cracking tools undo easily.
What should I use instead?
A password manager with long random passwords, or a passphrase of four or more random words, plus two-factor authentication. ~/secrets generates both.