Hardest/Impossible Composition Challenge?

follow-up #60:

For fun, I downloaded all rated, standard chess lichess games played in January 2017 (Lichess was smaller back then, with more manageable file size) and looked at the first 10 or so underpromotions to Bishop I could find.

As expected, all of them were jokes, with the game already decided and it didn't matter anymore... so if looking for interesting underpromotions in the Lichess dataset, I guess one needs to search either with a lot of patience, or filter with a script.

ply 92 joke
ply 93 joke
ply 122 joke

ply 137 joke
funnily enough, this one ended in a stalemate later in the game. The joke backfired badly...

ply 95 joke
ply 67 joke

ply 86 joke
underpromotion out of frustration, apparently...

ply 105 joke

ply 78 joke
A joke, but it has a certain style to it. A Bishop is enough to check the King, so why promote to a Queen?

ply 176 joke
Haha, how to mate with three bishops.

@Panagrellus

Once one understands the logical impossibility of advantageous R or B underpromotion not involving stalemate, empirical data is quite uninteresting, all the more if it is empirical data of amateurs, including beginners.

@nayf

I was not looking for a unicorn, and I'd expect that any meaningful underpromotions in the lichess dataset will have to do with stalemate one way or another.
However, as this thread has demonstrated, there are some interesting cases (e.g. the "defensive" self-stalemating and hooking up the stalemate threat with perpetual checks) .

There must be more examples like this in the lichess database, just because it is so huge: 889,835,852 rated games so far! The problem is how to find the needle in the haystack. It would be easy to identify all games with underpromotions, the tricky bit is to automatically identify those games where =R or =B is the best move...

By the way, a nice study for the "usual" reason (stalemate).

hmm underpromotion puzzles are slightly hard
some pawn endgames...

I discovered one can use the SCID database to filter millions of games for underpromotions.
This is a bit off-topic here though, so I started a new thread:

lichess.org/forum/general-chess-discussion/the-prevalence-of-underpromotion-to-rook-or-bishop#1