Visible characters are not the whole measurement
Unicode Passwords and Length: Count the Right Thing
A translator choosing a new password may see a short line of accented letters and emoji while a form reports a surprisingly large character count. Neither display necessarily tells the whole story. Text has a visible appearance and an encoded representation, and different measurements answer different questions. Before adding unusual characters to improve a meter, establish what is actually being counted, whether the candidate can be entered reliably, and what a language-limited estimator can reasonably recognize. A longer technical representation is not automatically a less predictable choice.
Use the Password Strength Checker with a proposed password, preferably an invented practice candidate rather than a real account secret. It accepts one masked Password input and reports code points and UTF-16 units after you select Check password. Those measurements describe the exact input; they do not test a provider account. This guide explains how to read them alongside an estimated score without confusing text length, portability, and resistance to guessing.
Separate visible clusters from the two reported counts
A grapheme cluster approximates what a reader experiences as one character. A code point identifies an element in Unicode text, while a UTF-16 code unit is a 16-bit element of the representation used by JavaScript strings. These are not interchangeable labels. For a small nonsecret illustration, an ordinary Latin letter occupies one code point and one UTF-16 unit; a character outside the Basic Multilingual Plane occupies one code point and two units. An emoji sequence can combine several code points into one visible cluster. The checker reports the latter two counts, not a grapheme total.
Read the field labels before comparing a count with another application. A destination that says only characters may be using a different measure, so a discrepancy is a question about its implementation, not evidence that your clipboard lost text. For the translator, the practical first step is to compare a harmless sample with the destination documentation. Do not send support staff the proposed secret to demonstrate the mismatch. A description such as one non-BMP character counted as two units communicates the issue without turning a private candidate into a troubleshooting attachment or requiring anyone else to handle it.
Masking dots are also a presentation choice, not a reliable length ruler. Do not count dots or assume that deleting one visible symbol always removes one code point. If inspecting a practice input helps, Reveal exposes the input and Hide masks it again. Use that view privately, particularly on a shared screen. The result does not provide a character-by-character map, matched-token display, or positions within the password. Keep your conclusion narrow: the supplied text has these reported lengths, not this many independent random choices and not this much guaranteed strength.
Keep exact input distinct from equivalent appearance
Some accented text can be represented by a precomposed code point or by a base letter followed by a combining mark. For example, U+00E9 and U+0065 U+0301 can render as the same accented letter while containing different code-point sequences. Unicode normalization defines ways of reconciling certain equivalent representations. That explains why matching appearances are not sufficient evidence of identical input. It does not mean every pair of similar-looking symbols is equivalent, nor that a reader should manually substitute characters inside a password until two counters agree.
This checker preserves every accepted character exactly: no trimming, normalization, or truncation. A leading or trailing space is not silently removed, and a decomposed accent is not silently converted to a composed form. It accepts single-line candidates only and rejects carriage returns and line feeds, including line-break paste or drop attempts. Check password evaluates an accepted candidate, not a cleaned version selected on your behalf. If you deliberately alter spacing or representation, you have made another candidate and must check it explicitly. The measurement cannot promise how a keyboard, another application, or a remote sign-in form will process similar-looking text.
A real provider can have its own documented handling. Current NIST verifier guidance discusses NFC normalization when Unicode is accepted; that does not turn this local candidate checker into a verifier or prove that every provider follows the same approach. For the translator, the useful question is whether the genuine account setup and later sign-in preserve a usable credential across the intended devices. If the answer is unclear, choose an independent candidate with a representation you can reproduce reliably. Avoid diagnosing an existing login by repeatedly changing accents in a live password field or sharing the secret with a support conversation.
Understand a rejected input before changing the candidate
The checker accepts a nonempty candidate within all three ceilings: 128 code points, 256 UTF-16 units, and 2,048 UTF-8 bytes. These are local processing limits, not recommended password targets or universal account requirements. UTF-8 bytes measure encoded data size, which is another quantity distinct from the two displayed length counts. Acceptance means the candidate fits this tool contract. It does not mean a destination will accept its length, spaces, script, or punctuation. Rejection likewise does not mean the candidate was found in a breach or identified as an attack; no such lookup takes place.
An unpaired surrogate is rejected rather than repaired or scored. UTF-16 uses paired surrogate units for characters outside the Basic Multilingual Plane; a lone unit is ill-formed text. This is different from rejecting an unfamiliar language or rejecting emoji in general. If a nonsecret sample triggers a malformed-text error, obtain a clean sample through the intended input method. Do not remove an arbitrary half of a character or accept a replacement glyph as proof that the original sequence survived. For a proposed credential, making a fresh independent candidate is often simpler than carrying uncertain copied text forward.
When a candidate exceeds a ceiling, the checker does not assess an accepted prefix and pretend the whole value was tested. There is no automatic shortening. Decide separately whether the destination needs a different format, then create a complete new candidate that fits both sets of constraints. A very long repeated slogan can still be predictable, while a technical limit says nothing about the wisdom of that slogan. If you are investigating the counters themselves, use repeatable public teaching text instead of an active secret. That lets you discuss the input boundary without disclosing a credential or preserving it in screenshots.
Measure before interpreting
Read the two length counts together
Check a practice candidate and distinguish exact text measurements from the model estimate.
Read multilingual scores with a limited model in mind
The result score runs from 0 to 4 and is an estimate of guessability under the bundled model. Its dictionaries cover English and common password material, not every language, local expression, or regional keyboard habit. A familiar saying in another language can therefore look less recognizable to the model than it is to a person who knows that language. Do not interpret unfamiliar-to-this-dictionary as unpredictable-to-an-attacker. The checker has no username, platform, or personal-context input, so it also cannot account for a candidate derived from the translator name, public portfolio, or well-known project title.
Pattern output consists of aggregate counts by type, not the matched words, their positions, or a reconstruction of the candidate. Generic static English feedback can indicate broad issues worth reconsidering, but it is not a translated linguistic analysis. A missing dictionary match does not establish that the candidate contains no dictionary words. Similarly, a repeat finding describes something the model recognized; it does not quantify every cultural association or personal clue. Use the feedback to question the construction process. If the candidate came from a quotation, public phrase, or predictable transformation, replace that process instead of shopping for a script that raises the score.
The model guess estimate is displayed as 10^estimated guess log when the base-10 logarithm is finite, or unavailable otherwise. For scale, a hypothetical exponent of six means an estimate on the order of one million guesses; it is not a timer, probability, or observed attack result. The checker displays no actual crack time. Two candidates with the same score can differ substantially in construction and suitability, and 4 does not mean secure or uncompromised. A random-generation process cannot be certified from one resulting string. Keep the score as one limited observation, not a certificate for an account or language.
Choose a candidate you can reproduce on the needed devices
Consider where the translator actually signs in: a desktop with a full keyboard, a phone with several language layouts, and perhaps a borrowed recovery device. An unusual symbol may be easy to select in one place and awkward to reproduce elsewhere. That is a usability question separate from the score. Before committing, review the genuine destination restrictions and the input methods you expect to use. A harmless sample can expose layout differences without risking an account lockout. Do not choose a shorter, personally meaningful phrase merely because it avoids one inconvenient key; choose a new candidate with both independence and practical entry in mind.
A password manager can reduce manual entry, but it cannot make an unsupported representation acceptable to a provider. Keep durable storage in a separate approved manager, not in this checker. The separate Password Generator can create a new candidate if you decide to start over. Moving between the tools is a manual handoff, with no password in a URL and no automatic transfer. The checker has no Copy button, generation function, vault integration, or export. Manual browser clipboard use is outside its control. Plan handling through your protected workflow rather than expecting an inspection result to become a stored credential.
For a new account, compatibility is established in the genuine provider flow, not by this meter. For an existing account, prefer assessing proposed replacements rather than entering its current secret here. If you already know the current value was exposed or reused, you do not need a favorable or unfavorable score to justify changing it. The check-and-replace guide explains that decision path. Separate the questions in your own notes: suitable representation, provider acceptance, independent choice, and completed account change. Recording those outcomes does not require recording the characters themselves.
Finish the inspection without relying on retained state
Load example fills the Password input but does not run a check. Select Check password when you want an estimate; typing or revealing the input is not a real-time scoring request. Clear removes the working candidate and derived result. Evaluation runs in a local worker that can be terminated, with a six-second limit. An unavailable result or interrupted job is not a low score and is not a warning about the destination account. Do not keep resubmitting a real secret to obtain a more satisfying display. Use a practice candidate when you need to understand how the controls behave.
The checker clears all input and derived DOM content on hidden visibility, pagehide, pageshow, and freeze. Switching tabs can therefore end the working inspection, and returning does not restore its candidate or feedback. There is no history, saved project, PDF, download, or account record to reopen. Arrange a short visible session and retain only the nonsecret decision if one is needed. A screenshot that includes revealed input defeats that small handling window. Even a score screenshot can be unnecessary when the meaningful decision is simply to discard a predictable proposal and create an unrelated replacement.
The genuine checker uses an opaque sandbox served by standalone PHP without a WordPress bootstrap, with no-store and connect-none restrictions. Local evaluation does not make normal WordPress page traffic disappear. It also does not protect against browser extensions, a compromised device, or a hostile parent page replacing the interface. Clearing is not secure memory erasure and does not erase your system clipboard or copies elsewhere. There is no breach database, account test, API, or AI analysis behind the result. End with a measured conclusion about this candidate, then use the provider and protected storage for the separate credential task.
Questions about Unicode Passwords and Length
Does one visible emoji always count as one character?
Not under every measure. A visible cluster can contain several code points, and a non-BMP code point occupies two UTF-16 units. This checker reports code points and UTF-16 units rather than a grapheme count. Compare the labels, not masking dots, and do not assume another service measures length the same way.
Will the checker normalize accents or remove outer spaces?
No. Accepted characters, including outer spaces, remain exact without trimming, normalization, or truncation. Candidates must be single-line: CR/LF and unpaired surrogates are rejected. A visually similar representation can be a different candidate. The provider may handle text differently; this result cannot prove compatibility or validate an existing login. Use nonsecret samples to investigate representation questions.
Does a high score prove a non-English phrase is a good choice?
No. The bundled English and common dictionaries are not comprehensive across languages. A familiar phrase outside that coverage can be overestimated. Consider how the candidate was chosen and whether it uses public or personal material. Prefer an independent replacement rather than translating a predictable phrase until the meter stops recognizing it.
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 the representation question specific
Review one proposed password
Use the checker for local candidate feedback, then handle acceptance and storage in their proper places.