@LukeShu note that json.Unmarshal documentation mentions this:

When unmarshaling quoted strings, invalid UTF-8 or invalid UTF-16 surrogate pairs are not treated as an error. Instead, they are replaced by the Unicode replacement character U+FFFD.

As json.Unmarshal and json.Valid share underlying code, your example looks like an expected behavior.

Perhaps there's a room for improving the documentation?

I understand that Unmarshal and Valid share underlying code, but Unmarshal being permissive of invalid input doesn't mean that Valid should identify it as valid (at least without its own note in the documentation).

And I discovered this because of inconsistency between the various functions, when I expected that even if they have quirks they'd at least be consistent because of sharing the underlying implementation. I'd discovered that json.Unmarshal(any)→json.MarshalIndent() would convert the binary garbage to U+FFFD, while json.Indent() and json.Compact() would pass it through unchanged. I figured that differing behavior for invalid JSON was reasonable, but then was surprised to find that json.Valid() identified it as valid. (And that differing behavior, while surprising, I couldn't really say was a bug; while json.Valid() identifying it as valid is a clear bug to me.)

CC @dsnet @mvdan

It's unfortunate that the json package allows invalid UTF-8 since RFC 8259, section 8.1 clearly states that JSON must be UTF-8.

Given that we already verbally promise handling of invalid UTF-8 in Unmarshal, we should also document how it operates in Valid.