NOCLAU GUIDE
Preserve Emoji and Multilingual Text During Cleanup
Compare cleaning modes without damaging joined emoji, connected scripts, or mixed-direction text.
A cleanup that looks harmless in an English identifier may change a multilingual sentence or an emoji. This guide is for editors who want to remove an accidental character without treating unfamiliar writing systems as noise. It explains how to build a small representative sample, compare conservative and custom behavior, and ask for an appropriate language review. The goal is to preserve the intended text while making only the changes needed for the destination.
Scroll the table sideways to see every column.
| Content | Default review | Avoid |
|---|---|---|
| Joined emoji | Preserve sequence; inspect destination | Deleting U+200D automatically |
| Connected scripts | Ask a fluent reviewer | Equating unfamiliar marks with errors |
| Mixed direction | Review punctuation and reading order | Bulk deletion of controls |
| Latin identifier | Check exact expected value | Applying that rule to every paragraph |
Start with representative content
Before processing a large document, collect a small sample containing every kind of content that matters. Include a plain identifier, a sentence in each writing system, any mixed-direction punctuation, and the emoji used in the final copy. Keep the sample beside the untouched source. If the destination is a mobile interface, use the same short labels and narrow widths that readers will actually see.
Do not replace a multilingual sample with Latin letters and assume the test remains equivalent. The role of a character depends on its neighbors. At the same time, a private document is rarely necessary for a bug report. Ask a fluent reviewer to help create an invented phrase with the same structure when you need to share a reproduction publicly.
Understand why one visible symbol can contain several characters
An emoji can be represented by a sequence of code points, and some sequences include a zero-width joiner. The Unicode emoji specification describes these sequences and their intended presentation. Removing part of a sequence may alter the displayed result. The precise appearance also depends on the fonts and platform, so inspect the actual destination instead of relying on a single browser preview.
noclau counts code points, not the number of visible symbols. For the constructed sequence woman plus U+200D plus laptop, the character counter sees three code points even when the system displays one joined emoji. The joiner is reported as a finding. That report describes the sequence; it is not a recommendation to remove the joiner.
Compare conservative and custom rules
Conservative mode preserves U+200C, U+200D, and U+2060. It also leaves supported special spaces and dash variants unchanged. It removes U+200B, U+FEFF, and U+00AD. This makes it a useful starting point for many accidental-separator problems, but it does not know which break opportunities or soft hyphens were intentionally used in a particular document.
Custom invisible-character removal can remove the joiners as well. Therefore the same text can produce different results under two modes without either action being a software malfunction. Read the rule note, inspect the proposed diff, and compare the result with the original. If the purpose of a mark is unknown, retain it and investigate instead of approving an edit because the count became smaller.
Run an emoji preservation experiment
Open the zero-width remover and enter a short sample containing an emoji joiner sequence and an accidental U+200B inside a separate Latin identifier. In conservative mode, the expected behavior is removal of U+200B while the U+200D remains. The output should preserve the emoji sequence and repair the identifier. This is a constructed acceptance test for the application's rules.
Now switch to Custom with Remove invisible selected and observe the proposed output before copying it. The joiner can be removed too. That comparison makes the tradeoff visible, so you can choose deliberately. Restore the original input after the experiment. Do not use the altered version in a real message until its appearance and meaning have been reviewed in the receiving application.
Give connected scripts a language review
Joiners and non-joiners may matter in connected-script text. A general character remover cannot judge the intended spelling or form from a string alone. If you are editing a language you do not read, ask a knowledgeable reviewer to compare the original and proposed result. Provide the exact code points removed, because two screenshots can conceal a difference in the underlying sequence.
The review should focus on actual words, not simply on whether the output appears neat. Ask whether the spelling, joining, punctuation, and line boundaries remain correct. Avoid stripping all non-ASCII characters to satisfy an unexplained validation error. If a field rejects legitimate text, review the field requirements and seek a compatible representation rather than silently changing someone's name.
Preserve direction until you understand the layout
Mixed left-to-right and right-to-left text can need careful punctuation and number placement. noclau reports selected directional controls but does not delete them automatically. That behavior helps preserve the sample while you investigate. It does not certify that the display is correct or replace the need for direction-aware layout testing.
Check the paragraph in the actual destination, including link labels, brackets, and punctuation near language boundaries. A text input and a published HTML paragraph may behave differently. If a control is unnecessary, make a targeted edit in a suitable editor and repeat the comparison. Record why the change was made so a later editor does not reintroduce it or remove other marks by analogy.
Test more than the desktop screenshot
Use the intended application on the platforms your audience relies on. Compare the saved text as well as its rendered appearance. A font substitution may change an emoji's appearance without changing the stored sequence, while a copying step may change the sequence itself. Those are different problems and should not be reported as a single generic cleanup failure.
For layout-sensitive content, check long labels, wrapping around numbers, and narrow screens. For data fields, repeat the relevant matching or validation operation. Preserve the original throughout review and accept a result only when the concrete problem is resolved without a new language or display issue. The most appropriate final output may still contain detected characters because those characters belong in the text.
References and further reading
Examples on this page are constructed demonstrations, not measurements of how often a platform inserts characters.