I like that you want to separate facts, and "social interaction". Facts cannot be debated; social interaction shouldn't, because there's no accounting for taste (to each their own).
Some suggestions for the facts:
When blackpowder was introduced on BGA, it was referred to as P (powder). Using K (blacK) came over from users who also played on Hanabi.live. Using K seems just as weird to me as using E or L for blue. It also works in French:
R = red/rouge;
Y = yellow/yaune;
G = green/vert;
B = blue/bleu;
W = white/blanc (there's already a comment that blanc/bleu both starts with B, hence English notation is used);
M = multicoloured/multicolore
P = powder/poudre.
Conventional Leftism differs from BGA standard: it allows for finesses, and it changes the focus of a clue to the left-most-logical card, instead of the newest unmarked card. (Sometimes called: 1-clue-1-play.) Example:
Bob has (new --> old) B5, G4, W3, Y1, B2.
The B2 was saved earlier; it is marked with 2. B2 is now playable, since B1 is on the table.
In CL, Alice can give blue-clue to mean: play B2, save B5. In Standard blue-clue would mean: play newest blue-card (it is B3) after playing B2.
Move 1.8 Conventions to the tips & tricks. The conventions, and definitions are not fact. Examples:
"Marked" is used on BGA, "Touched" is/was used on hanabi.live. "Early save" is used by some players to indicate a save-clue that saves a card that isn't in danger of being discarded (often used to avoid complications due to convention restrictions); but some other players use the term to describe saving non-unique cards. "Non-unique" is mostly used on BGA, hanabi.live use(s/d) "critical".
You can include in the fact section:
According to this open source program
https://github.com/Quuxplusone/Hanabi
you can get perfect score in the standard (5-colour) variant in 94%, 98%, 98%, and 96% of the decks for 2P, 3P, 4P, and 5P games respectively (when cheating / using perfect information).
Going to the tips & tricks:
I love that you start with the maths as an introduction to "why conventions are needed". However, I would not use the last two lines, where you calculate cluetokens / card. It's more useful to think of 22 tokens : 25 card = -3 (so you're three tokens short, and need to make at least three tokens count double); or 39 tokens : 35 cards = +4 (so you can use four tokens for stalling/re-cluing).
I would follow up with a strategy section. If you assume you want to play APARP (as perfect as reasonably possible), you should start with the assumption that you can mimic cheat-bot (see facts) and get perfect score pretty much all the time. Dropping a "first" card gives you a chance that the "second" card is at the bottom of the deck, and it will prevent you from getting perfect score. In addition, if a second 3 is one card above the bottom of the deck, there is a 50-50 chance that the order of the 4 and 5 is "bad" (Alice draws 3; Bob has 4 and 5 vs Alice draws 3 and has 5; Bob has 4). For other than 2player variants, the calculations are more complex, and you need to factor in excess clue-tokens (for stall clues), but as a rule of thumb you can say that it's roughly 50-50. This also goes for lost 2s with the two cards above the bottom deck. WIth this approximation, you can calculate a "penalty score" X:
+1 for every lost first 4;
+1.5 for every lost first 3;
+2.5 for every lost first 2.
The chance that you lose due to "bottom deck" is X /N (where N is number of cards remaining in the deck). This means that the primary choice a team should make, is whether to aim for "guaranteed" perfect score, or "likely" perfect score.
After the strategy choice, comes convention. Convention tells you what meta-information you assign to the clues. This is completely arbitrary, and unfortunately the source of a lot of toxicity. This player
https://boardgamearena.com/player?id=83880676 once proposed something extremely-meta: Use red-clue to indicate play position 1; use yellow-clue to indicate play-position 2; etc. Use 1s-clue to save 1 card; use 2s-clue to save 2 cards, etc. People would call this "cheating", and "it wasn't logic", but this meta-information is just as arbitrary as the assumption that a marked card is useful (junk cards aren't marked).
As a final word, which is a second source for toxicity: ELO. Using the ELO tables as described, removes the correlation between skill and ELO. Not getting perfect score should, in my opinion, be assigned negative ELO-points, since perfect score is almost always possible. In addition, the tables that are linked imply that it is easier to get perfect score in 2player mode, while this is the hardest variant to play (because the team members have the least amount of information). Also, the "elasticity factor" for ELO is extremely low. Since you can get a maximum of +10, or -10 points, it takes >70 perfect games to become master, or >50 games for a rank-inflater (~1200 ELO) to drop below master (master = 700 ELO). Since you can filter tables on ELO (and not on skill), you want there to be a correlation between skill and ELO, and preferably one that matches the (arbitrary) levels set by BGA (100, 200, 300, 500, 700). I've recommended to cap ELO for hanabi (and other cooperative games) at 1000 points, where 1000 points should reflect: has gotten perfect score in the last 20 - 30 games. (100.0%). (And 700 points means has gotten perfect score in 70.0% of the last 20 - 30 games.) Perhaps this can somehow be addressed in the wiki?