I considered that, but wanted to preserve Crockford's partial defense against accidental profanity generation while still allowing U/u to be decoded (if Crockford's aliased it I probably wouldn't have bothered to whip up Base32H, lol).

As for 1/l/I, yeah, that's definitely the main flaw. The workaround (as described in the FAQ) would be to always emit uppercase and take advantage of L being pretty visually distinct. A bit ugly for URLs, but for things like asset tags, product keys, and item codes / SKUs (the main things I had in mind) that's already the norm.