A summary:

An old proposal (2015, not sure why OP posted it now), that basically proposes to put some more standards and limitations around JSON formatting to make it more predictable. Most of it seems pretty reasonable:

• Must be UTF-8 encoded and properly escape Unicode characters
• Numbers must respect the JavaScript number Type and it's limitations (i.e. max magnitude of an int etc.)
• Objects can't have duplicate keys
• The order of keys in objects does not matter
• A JSON file does not need to have a top level object or array, it can be any JSON value (i.e. just a string or a number is still valid JSON).
• It proposes that when processing JSON, any unrecognized keys should be ignored rather than errored

It recommends:

• Specific formats for date-time data
• That binary data be stored as a bas64url string

Honestly, the only part of this I dislike is the order of keys not mattering. I get that in a bunch of languages they use dictionary objects that don't preserve order, but backend languages have a lot more headroom to adapt and create objects that can, vs making a JavaScript thread loop over an object an extra time to reorder it every time it receives data.

Why restrict to 54-bit signed integers? Is there some common language I'm not thinking of that has this as its limit?

Edit: Found it myself, it's the range where you can store an integer in a double precision float without error. I suppose that makes sense for maximum compatibility, but feels gross if we're already identifying value types. I don't come from a web-dev/js background, though, so maybe it makes more sense there.

Just skimmed but seems like a decent idea. Not that I've knowingly run into issues parsing JSON too much