Character reference expands to a control character (U+X).

A numeric character reference such as  or  resolves to an invisible control character, and HTML doesn't allow those in a document.

Control characters are the non-printing codes U+0001 to U+001F and U+007F, left over from teleprinters and terminals. U+0003 is "End of Text" and U+001B is "Escape", for example. The HTML parser reports an error for any character reference that resolves to one of them, in text or in an attribute value, and the validator names the code point in lowercase hex: (U+0003), (U+001b), (U+007f). Three of them are whitespace and stay valid: 	 (tab), 
 (line feed) and  (form feed). �, 
 and references from 128 to 159 get their own messages.

These references rarely come from hand-written markup. They usually come from text pasted from another application or imported from a database, where an invisible character sat between two words, and a CMS or template then escaped it as  or similar. The character isn't a space, so the words on either side can run together, and some browsers draw a placeholder box where it sits.

Replace each reference with the character that was meant, usually a plain space, or delete it. When the text comes from a database or a feed, clean the stored text or strip control characters before it reaches the template, or every page built from it will repeat the error.

Invalid example

<p>The workshop starts at&#3;9:00 and ends at&#x1b;13:00.</p>

Valid example

<p>The workshop starts at 9:00 and ends at 13:00.</p>

Learn more:

MDN Reference

MDN - Character reference

→

WHATWG Spec

HTML Standard - Parsing HTML documents

→

WHATWG Spec

Infra Standard - Code points

→