Understand the estimate before acting on it

Password Strength Scores Explained

A password score answers a limited question: how difficult does this candidate look to a particular pattern model? It does not answer whether an account is protected. Read the score together with the model guesses, length fields, and pattern categories, while keeping the questions the model cannot answer separate.

Use Password Strength Checker with a public example. Load example fills the masked Password candidate field only; choose Check password to run the estimate. Learn what each result means before considering any disposable candidate of your own.

Treat 0 to 4 as categories rather than grades

The checker reports an estimated score from 0 to 4, giving five categories rather than a continuous measure of security. The categories group model estimates into broad ranges. A higher category means more estimated guessing work within that model; it does not mean every part of a password is good. Do not read 2 out of 4 as fifty percent secure or 4 out of 4 as a perfect result. There is no meaningful percentage of account safety hidden behind the display. The compact score is a way to summarize a calculation, and summarizing necessarily leaves out detail that may matter to your decision.

Category boundaries can make a small edit look dramatic. If one public example lands near a boundary, a slight change may move its score up a whole step. Another change can increase the guess estimate while leaving the category unchanged. Neither outcome is a reason to tune a real secret until the display reaches a target. Compare the construction and the supporting fields instead. Ask whether the change removed a predictable habit or simply crossed a cutoff in this particular calculation. A score increase is evidence about the model response to that exact input, not a measured improvement in the security of an account.

The top category is especially easy to overread. Many different candidates can share it while differing in length, origin, and real-world exposure. The model does not know whether one was copied from a public article, reused for several accounts, or sent to someone in a message. A published example is unsuitable as a secret even if its category is high. Likewise, a low score is not proof that an attacker has already recovered anything. Separate the useful warning about a predictable idea from any unsupported story about a current compromise. The result contains no evidence about the history or ownership of the supplied candidate.

Understand model guesses without inventing a deadline

The local estimator looks for familiar constructions and builds an explanation for the whole candidate. Model-estimated guesses describe the guessing work associated with that explanation. Think of the figure as an informed estimate of search order, not the size of every possible string with the same length. A person who picks a familiar word and predictable suffix has not chosen uniformly from all combinations of letters and symbols. That is why a pattern-based estimate can differ sharply from a simple character-count calculation. The checker is inspecting supplied text with a model; it is not attempting guesses against a provider or checking whether a guessed value logs in.

The guess display uses 10^L, with L representing the model-estimated base-10 guess logarithm. It appears only when that returned model value is valid. A missing or invalid value does not mean zero guesses, infinite strength, or an automatic top score. Large exponents also need modest interpretation: software can format a value precisely while the underlying assumptions remain approximate. Avoid quoting decimal places as though they came from a physical measurement. The important distinction is between a calculation result and a guaranteed lower bound on an attacker. This tool provides the former and does not establish the latter; the notation is not an actual attack counter.

There is no actual crack time in the result. A real duration would depend on conditions this checker does not measure, including how a service limits attempts and how stolen password data was protected. Nor does the checker claim to measure entropy in a human-chosen string. That requires information about how choices were made, not just what the final text looks like. Do not turn the displayed guesses into a promise of years, a deadline for an attacker, or a number of security bits. The useful conclusion is smaller: a local pattern model estimated this amount of guessing work for this exact candidate under its own assumptions.

Read length and pattern counts as different evidence

Length fields describe the text, while pattern fields describe parts of the model explanation. The Unicode code-point count and UTF-16 code-unit count may agree for ordinary letters and differ for other characters. A symbol can use two UTF-16 units, and a single visible character can be formed from multiple code points. Neither field measures how many decisions a person made when choosing the candidate. A long repeated construction can still be predictable. Conversely, a visually short string can use several units without gaining the independence of several random choices. Read length as a concrete property, not a shortcut to a strength claim.

Pattern TYPE counts deliberately avoid the actual matched tokens, positions, and secret segments. A dictionary type may indicate familiar words or common-password material within the selected explanation. Other types can describe repeats, sequences, dates, or keyboard movement. A count is not a list of private fragments and should not be used to reconstruct one. It also does not identify every possible pattern a different system might find. The estimator chooses among explanations that can overlap. A change in one type count may reflect a different whole-string explanation rather than the disappearance of every recognizable feature of that kind from the candidate.

Do not rank candidates by the fewest pattern types or treat zero in a category as a certificate of absence. Some parts may simply be handled as unmatched text by the model. Similarly, many short recognized pieces are not automatically worse than one long familiar phrase; the estimate depends on how the pieces are weighted and combined. Use counts to ask sensible construction questions, such as whether repetition still plays a role after an edit. Keep that lesson general. The result is intentionally less revealing than a highlighted password breakdown, so it cannot tell you which exact substring to replace or where a private segment begins.

Read beyond the number

Inspect the supporting fields

Check one public candidate and read the score alongside valid model guesses, lengths, and pattern types.

Allow for dictionary and personal-context blind spots

The bundled dictionaries focus on English and common material. They are not a live catalogue of every language, current phrase, local reference, or password someone might choose. A quiet dictionary summary can mean the model lacks relevant coverage, not that a phrase is obscure. This matters when the words are familiar to a particular community or when a new expression becomes popular after the bundled data was assembled. Do not interpret a high estimate for unfamiliar-language text as proof that the language itself makes a password stronger. The uncertainty belongs to the model, and the person choosing a candidate still needs to consider its origin.

Personal knowledge creates another gap. A stranger and a colleague may have very different starting information about someone. The checker does not ask for usernames, platform names, birthdays, email addresses, or other context, and you should not add such information to the candidate field. It is not a form for building a profile of an account holder. If a candidate relies on a personal reference that other people know, the score cannot make that reference private. Make that judgment outside the tool without entering more information. A result based only on candidate text cannot establish resistance to guesses informed by your life or work.

Common-password recognition is also not a breach search. Bundled dictionary material helps estimate familiar choices, but the checker does not query a current breach service or determine whether the full candidate appears in a leak. Absence of a finding therefore cannot establish that it is uncompromised. A candidate already shared publicly remains exposed regardless of whether a dictionary happens to recognize it. Separate those facts when explaining a result to someone else. Say that the local model returned a category, not that it checked all known leaks. There is no remote API or AI analysis to fill these gaps, and no account identifier is collected for a targeted assessment.

Turn generic feedback into a useful choice

Warnings and suggestions are generic, static English text. They do not quote secret pieces or provide a tailored rewrite of the candidate. That restraint keeps the output focused on a pattern lesson. Treat advice about familiar choices or predictable edits as a reason to reconsider the recipe, rather than a command to append one more symbol. An interface can help you notice a habit, but it cannot prove that a personal variation is now unknown to everyone else. When advice seems broad, return to the construction question: what would make this idea easy to anticipate before anyone saw the exact text?

For a new candidate, a separate generation process is more useful than repeatedly decorating a familiar base to reach the top category. The Password Generator is a separate tool, not a hidden checker mode. Opening it does not transfer the current candidate in a URL or generate a replacement inside this session. Keep any fresh secret out of shared demonstrations and public notes. This guide does not ask you to disclose an existing password to prove that it should be changed. If the real issue is known exposure or reuse, deal with that issue through the trusted service workflow rather than seeking reassurance from a score.

A candidate also needs to be manageable in the setting where it will be used. The checker cannot confirm that a service accepts its characters, that a particular mobile keyboard can reproduce it, or that account recovery is set up correctly. The separate 2FA Authenticator handles a different kind of value; password candidates do not belong in an enrollment-secret field. Likewise, a strength estimate cannot enable an additional factor or validate a one-time code. For a structured lesson, use the sequential public-example comparison to explore one construction change at a time without keeping a collection of real secrets.

Keep the estimate inside its working boundary

Choose Check password explicitly; typing does not run live analysis. Load example fills only, and Reveal or Hide changes visibility rather than scoring. Candidates must be single-line. CR and LF line breaks are rejected, including in paste or drop input, not removed to produce another string. Accepted characters, including leading and trailing spaces, are preserved without trimming, normalization, or truncation. The maximum is 128 Unicode code points, 256 UTF-16 units, and 2 KiB of UTF-8 data. These bounds define this local task, not another service. Rejected input has no supported estimate here; rejection is not a strength verdict.

The genuine isolated PHP document has an opaque origin, no WordPress bootstrap, a no-store response, and connect-src 'none'. Its local worker is terminable and limited to six seconds per job; a timeout is not evidence of attack resistance. Clear removes input and all derived DOM content and stops the job. The same wipe and termination happen when hidden and on pagehide, pageshow, and freeze. Switching away is therefore an end to the working session, not a way to park a result. There is no history, persistent storage, autosave, copy, export, PDF, or password-generation feature in the checker.

Those boundaries do not promise secure memory erasure or protection from extensions, a compromised device, or a hostile parent replacing the genuine interface. Manual browser clipboard actions are outside tool control; cleanup cannot erase clipboard contents. The surrounding WordPress page has normal traffic separate from candidate analysis. Prefer public examples over real existing secrets, even on a trusted device. Paid SMTasker supports nine-platform social automation with Skye and MCP, not password custody or security certification. The checker neither updates provider logins nor supplies candidate text to that workflow. Finish narrowly: the model returned an estimate for a disposable candidate, while wider account-security questions remain separate.

Questions about Password Strength Scores

Why can two very different candidates both score 4?

The score is a broad category, not a unique rating for every candidate. Different estimates can fall in the same category. The top category proves neither uniqueness nor absence from breaches, and public examples must never become real secrets.

Can I convert model guesses into years or entropy bits?

Not as a reliable measurement of this candidate. The checker observes neither a real attack rate nor the process used to choose human text. It deliberately provides no actual crack-time or entropy claim.

Does no dictionary finding mean the candidate is unknown?

No. The model uses English and common dictionaries, not complete language coverage, personal context, or a current breach lookup. A missing finding is only an absence from the selected model explanation, not proof that nobody would recognize the construction.

Continue with another password-checking guide

Check a proposed password in the Password Strength Checker or choose a related task below. Explore more practical utilities in all free tools.

Keep conclusions proportional

Use the estimate as a learning aid

Return to one candidate at a time. A category is model feedback, not a safety certificate for an account.