What's the status on this? There's been a lot of cleanup recently, we should survey the document on whether this has been fixed.

Section titles can be checked using some ungodly regexes, so I'll see to this this evening.

sed -E 's,^.*<h[1-6]\s+id="[^"]+"><a(\s+[^=>]+="[^"]*")*>([^<]*)</a>.*$,\2,;t;d' docs/pandocs/print.html

Another inconsistency we have is (R) vs. (Read Only), and (R/W) vs. (Read/Write).

I suggest (Read-only) and (Read/Write) respectively, for clarity.

FF40 - LCDC - LCD control (read/write) seems good enough to me. No capitalization in the parentheses, as they are not beginning a new sentence.

Since we're changing all reg descriptions, I'll say that I'm fond of neither current style (ADDR - Name - CGB only - Description (R/W), ADDR - Name (Description) (R/W) - CGB only). I'd prefer e.g. ADDR - Name: Description (R/W) [CGB only], since it's imo more structured.

Additionally, we need to set a rule for when a register is partially readable and partially writable (or both). The current style seems to be not to mention the property at all, and I agree with it.

Another note: make sure to correctly differentiate "CGB only" (= the register is writable even in compat mode) and "CGB Mode only"; this should also be defined somewhere. (At this point, I think that #194 wouldn't be enough, and we would additionally need a "Conventions" page...)

If a register has unused bits, but all the usable bits have the same accessibility (e.g., SVBK), it should be labelled by the usable bits. If usable bits differ in accessibility (e.g., P1/JOYP), not mentioning it at all should be okay, since the list of bits will need accessibility indicators anyway.

As for CGB registers, that could be marked in the same parenthetical to avoid bracket overload: (read/write, CGB mode only).

Checked a few datasheets, most do not mention R/W in the register title at all, and only per bit.

Examples:

That's reasonable when registers are mostly bitfields, but the GB has many that aren't. For instance, there's no reason to specify a per-bit accessibility of SCX.
And if you have to specify byte-wide accessibility in some cases, why not do it whenever it is reasonable?

Given that we don't specify bits individually, but whole bitfields at once, this would be less of a burden.