1. One source of false positives I can think of would be if the pattern [ s ] appears elsewhere, such as in pushed arbitrary data (OP_RETURN). From a purely statistical perspective, the chance is virtually 0. Unless some people and/or miners use such for something (maybe Layer 2 stuff, or simply to f* around). I hypothesize false positives on the actual mainnet blockchain are exceptionally rare. Still, I haven't taken the time to quantify the exact number (and I would be surprised if it wasn't very, very close to or exactly 0).

2. The blk files simply the blocks as they come, the tx data bundled into ~130 MB "units" for convenience/standard, and saved as raw bytecode to be space-efficient (using all 256 bits (0x00-0xFF) instead of limiting itself to the number of an alphabet (hexadecimal is only 16 "letters", duh!). That's why they look funny, and maybe get you to think "this gotta be some weird code", if you open them as they are without first converting to something human-readable.

--snip--