Telegram error 400: Bad Request: can't parse entities

You set parse_mode (MarkdownV2, Markdown or HTML) and the text does not parse under those rules. The description always begins can't parse entities: and then continues with a parser-specific detail that varies by input, for example Can't find end of the entity starting at byte offset 42, Unsupported start tag "br" at byte offset 10, or Character '.' is reserved and must be escaped with the preceding '\'. Treat the prefix as the stable part and the remainder as diagnostic.

Nothing was sent. Telegram rejects the whole message rather than sending it unformatted. The byte offset refers to the UTF-8 byte position in the text (or caption) field you submitted, so multi-byte characters before the failure shift the number.

The MarkdownV2 escaping rules are documented exactly, and they are stricter than most developers expect. Outside of entities, all of these characters must be escaped with a preceding \: _ * [ ] ( ) ~ ` > # + - = | { } . ! — eighteen characters, including the period and the exclamation mark. Inside pre and code entities, only ` and \ must be escaped. Inside the (...) part of an inline link or custom emoji definition, only ) and \ must be escaped. Any character with code 1-126 may be escaped anywhere, so over-escaping ASCII is always safe. HTML mode instead requires &, < and > to be replaced with entities and accepts only the documented tag set (no <br>, <p> or <div>).

Libraries do not pre-validate for you: python-telegram-bot raises telegram.error.BadRequest and aiogram v3 raises TelegramBadRequest with the description; both ship escape helpers (telegram.helpers.escape_markdown(version=2), aiogram.utils formatting tools) that solve it at the source. This error is deterministic: the same input always fails, which is why it looks intermittent in production — it depends on what the user typed or what your database returned.

{"ok":false,"error_code":400,"description":"Bad Request: can't parse entities: Can't find end of the entity starting at byte offset 42"}

Never concatenate raw strings into a formatted template. Use a builder that escapes per parse mode, or send plain text with an entities array so there is nothing to parse. Add a unit test that sends representative user-like strings (underscores, dots, angle brackets, emoji) through your formatter and asserts a round-trip with the API's rules. Character budgets for the text being formatted are under text and caption limits; the production failure pattern is cause 7 in Telegram bot not responding.