Catch every mistake, not just the misspelled ones

The mistakes a spell-checker structurally cannot catch

A dictionary check asks one question of every word: is this string in the word list? Every error below passes it.

• Agreement between subject and verb. “The list of open issues are attached” → “is attached.” The verb agrees with list, not with the noun beside it.
• Tense and aspect. “I work here since 2019” → “I have been working here since 2019.” Since forces the perfect; the word work cannot.
• Articles. “She is manager of the team” → “She is the manager.” Articles depend on what the reader already knows, and no word list holds that.
• Homophones. “The company lost it's biggest client” → “its biggest client.” Their, there and they're; your and you're; affect and effect.
• Comma splices. “The build passed, we shipped it” → “The build passed, so we shipped it.” Two independent clauses glued with a comma.
• Prepositions and collocations. “responsible of the release” → “responsible for the release”; “discuss about the plan” → “discuss the plan.” Learned pairings, not rules you can derive.
• Punctuation that moves the meaning. “Let's eat, Grandma” and “Let's eat Grandma” are both perfectly spelled, and only one of them is an invitation.

Four errors, zero misspellings

You wrote

Their going to send they're feedback by Friday, its already late.

FixMeBot returns

They're going to send their feedback by Friday; it's already late.

It corrects. It does not re-voice you.

Correction changes what is wrong and leaves what is merely unusual. That is a narrower job than “improve this text”.

• Your contractions stay contractions: it never decides that “don't” should be “do not”.
• Short sentences stay short, and a deliberate fragment in a headline stays a fragment.
• Domain terms, product names and internal jargon are left alone, not “corrected” into everyday words.
• The order of your points does not move: a wrong order is an edit, not an error.

When the sentence should sound different, Paraphrasing sets the register and Length Adjustment sets the size. Correct first and re-voice second, because that order costs the least rework.

Why the unit is the sentence, not the word

Most of the errors above have more than one valid repair, and only the surrounding words choose between them.

• “I have sent it yesterday” becomes “I sent it yesterday” or “I have already sent it.” Yesterday is a finished time, so the past simple wins.
• “There ready” is they're ready about people and there, ready about a place; the two words in front settle it.
• “The data is” and “the data are” are both defensible; which counts as a mistake depends on the document's register.

Which is why a fragment corrects badly and a paragraph corrects well. Send whole sentences, and the one before the sentence you are unsure about.

Writing in a second language is a different error profile

Native and non-native writers make different mistakes, and only one set is visible to the tools already on your machine. Typos get underlined; articles, prepositions, aspect and word order do not, because every word is spelled right.

Correction runs in every language FixMeBot supports, and each has its own version of the trap: German case endings after a preposition, Russian verb aspect, Spanish ser against estar, French participle agreement, Polish and Czech declension. None is a spelling question, and all of them are why a text reads as foreign with no misspelled word in it. The full list is on the languages page.