How is this game drawn by insufficient material?

https://www.chess.com/game/live/19260455965

I only have a bishop left and he has three pawns. He lost on time but it said that the game was drawn by insufficient material. However, it is possible for Black to checkmate White as White can promote one of his pawns to a knight and get into this position (for example):

It's the same idea with having two knights, the mate isn't forced so no. That's my understanding at least.

I know that the mate is not forced, but it is possible to get mated as White. So technically White should lose as he lost on time with enough mating material left for Black.

So in B vs Knight where you have the bishop if you flag your opponent you win because mate is possible? I don't think that's right

Draw because white wouldn't march his king to a corner and underpromote to get mated. 

The site does not use the FIDE implementation of mate possible by any series of legal moves and uses something closer to the US Chess implementation, with some exceptions. Basically, only the material the side with time is looked at, disregarding all other material. So, if that isn't enough to mate on it's own against a lone king, it's a draw.

https://support.chess.com/article/268-my-opponent-ran-out-of-time-why-was-it-a-draw

In chess.com, it's a draw as long as you have at most one minor piece and no pawns, regardless of what your opponent has. Even if there's a sequence of legal moves that leads to checkmate, if there is no sequence of legal moves to checkmate against the lone king with the material you have, then your material is considered insufficient. Note that 2 knights would not be insufficient material in chess.com - you would win if you had 2 knights when your opponent flagged.

In FIDE, you win as long as you could checkmate with any sequence of legal moves. This gives the most winning chances. I believe that minor piece vs any pawn on your opponent's side is a win (they can underpromote to the other minor piece and you capture everything else), and minor piece vs any minor piece other than bishops of the same color is a win. (Minor piece vs only major pieces left and no pawns is a draw). There are certain exceptions such as if your opponent is forced to checkmate you which makes this harder to implement in chess.com. See the last problem here where despite black having a queen, the only sequence of legal moves leads to white checkmating black.

In USCF, you win as long as you could force checkmate. If your opponent has a lone king and flags vs your 2 knights, it's a draw in USCF but a win on chess.com.

Therefore, if you get in a situation where you can force Stamma's mate, your opponent can simply stall and it will be a draw in chess.com, while it would be a win in FIDE and USCF (and definitely anyone would agree this should be a win). I agree with the FIDE interpretation; if you time out, it should be assumed you play the worst possible moves from then on.

One solution (because it's so hard for a computer to figure out when helpmate is possible) is to do a quick search (programming an attempt to capture all of your opponent's pieces if you have a major piece or pawn left, or capturing everything except one minor piece if you have a minor piece left, is probably easier than programming stockfish), and if the search fails then the result is declared as a draw for a certain amount of time after the game ends (say, 30 minutes for live and whatever the time control was for daily), give the option for the side that didn't flag try to play out the helpmate themselves (and they earn the win if they manage it; they can restart as many times as they want). Since these cases are pretty rare, it shouldn't take too much time.

Interesting, that must be a fairly recent change. Which is weird, since the reason given is true for pretty much any other combination of material (i.e. the FIDE implementation)

I also just realized that the in the last problem of this article that I linked to earlier black can win by flagging white on chess.com, but not on USCF or FIDE, despite FIDE generally giving the player who flags his opponent better winning chances (FIDE is however strictly pareto better for the side the flags the opponent than USCF; in every position that is a win in USCF, it's also a win in FIDE).

In the 2 knights thing, you could checkmate the bare king though it can't be forced. Chess.com checks if you are able to checkmate against the lone king with the material you have left with any sequence of moves and with any starting position of your material, and gives you the win if you could.

That's rather depressing for me

I believe it is because you couldn't use a forced mate. With that said I see you are a CM so I am not sure if you are asking it as a trivia question or if you are actually asking for your knowledge. 

You're so stupid to not play e5 xD

Not all the time. I forgot the link to one forum thread by another player, but in the game Chess.com awarded a win for the player having two bishops residing on the same-coloured squares. There were only two kings besides the two bishops by the same player (KBB vs K), so there was no possible checkmate sequence but the side having the bare king lost on time and the other player won on time. 

That was quite a while back, but I am not sure if this has been updated as of now.

EDIT: Found the link https://www.chess.com/forum/view/general/why-isn-t-this-insufficient-material

It's like the case between Alireza Firouzja and Magnus Carlsen in the blitz game: Alireza lost on time and he lost even though Carlsen only had a bishop left...

If you run out of time on an equal position (both players can checkmate each other but the game should be a draw with correct play) you lose on time. It doesn't matter that there was no forced checkmate. If there was a way for your opponent to win, even though it is not forced, you lose.

While black lost on time, white could not have checkmated black so under these rules, it was a draw.

FIDE rules:  a loss because there is a potential helpmate.

US Chess rules: a draw because, looking at the unflagged player's material, it is one of the few cases (K+2N with no pawns, K+B, K+N) where a forced mate is needed instead of a helpmate (if there is a forced mate then it is still a win in US Chess).  When the unflagged player has a minimum of K+P or K+R or K+Q or K+2B (opposite color squares) or K+B+N or K+2N (when the opponent has at least one pawn) are all wins in US Chess (in normal play - blitz has more draws) regardless of position.

Chess.com: draw because K+B vs K is a draw and the flagged players' pieces (other than the King) are removed from consideration.

For that matter, White Ke1, Pa4, Pc4, Pf4, Ph4 vs Black Ke8, Pa5, Pc5, Pf5, Ph5 is s draw in FIDE and US Chess but would be a win on time in Chess.com (for the opponent of whichever player flags first) because ignoring the flagged player's pawns gives a K+4P vs K calculation and any one of the Pawns could become a queen.

In the above position in post #1, it's White who loses on time.

Either way, both sides are able to launch checkmates, but this site does not recognise having a lone bishop and a king as "being capable" of launching a checkmate, so it's a draw for Black.

I think that it would be difficult for Chess.com to convert the scenario to a win because the extra programming needed would, in my opinion, be a complex matter.

To implement a portion of the FIDE rules the following could be done for the unflagged player:

K+P or K+R or K+Q or K+2B (different color squares) or K+B+N or K+2N - award a win.  Material in addition to what is listed would not change it.

K+B vs (K+P or K+N or K+B on different colored squares) is a win.

K+N vs (K+P or K+N or K+R or K+B) is a win.

That would award all of the wins that FIDE awards.  However it would also award wins where FIDE (and US Chess) would award draws - when there is no legal sequence of moves to allow a helpmate.  Since Chess.com is already awarding such "wins", that may or may not be seen as a big deal.  Matching the FIDE rules and awarding draws in those situations would require a lot more complexity in programming.