Mate in one considered as a draw

@3
Yes, this is an incorrect implementation on chess.com.

This position for example is a forced checkmate for white,
but chess.com erroneously calls it a draw if black lets his time run out.

This position for example is an immediate draw by dead position,
but chess.com calls it a loss for whoever runs out of time.

wow that sure would b eannoying if you got one of hese in a game

Correct.

Chess.com's GUI doesn't correctly implement FIDE's dead position rule.

But who is to blame in that case?

It's because FIDE doesn't suggest any practicable way of determining whether a position is dead according to their rule in the general case and nobody has managed to produce any such practicable way. There are no GUI's or engines that can reliably recognise positions that are dead according to the rule. (Indeed there are also arbiters who refuse to recognise positions that are dead under the rule in conjunction with the 75 move rule and, possibly, also the 5-fold repetition rule.)

I would say the fault lies with FIDE in introducing the rule in the first place. To state a rule without giving a practicable way to check if the rule applies is nonsense.

The rule should be rescinded. Optionally, extra rules could be introduced for verifiable cases, possibly including your second case. What should happen in your second case, if not covered by such an extra rule, is that the sides agree a draw under rule 5.2.3.

I would just play Queen to the first rank and mate in 2 that way. But you are right. This site doesn't implement mating material and flagging correctly. It automatically declares certain positions drawn or lost based on only one side of the mating material aspect. If white has a king and knight and black has other pieces, even if white can force a mate with the knight, it declares it insufficient mating material. So in a blitz game, if white sets this trap:

Immediately after bxa7 it's declared a draw, even though both sides can theoretically mate each other and white has a mate in 1.

@8

"who is to blame in that case?"
++ The programmers.
We now have artificial intelligence that outperforms humans in many tasks including chess,
but the programs are unable to detect a dead position,
which any human arbiter can decide at a glance.

@9

It is too bad that on chess.com you can get away with a draw in face of a forced checkmate by letting your clock run out of time.

The problem is dead positions can be absurdly complicated, yet still drawn, such as this one:

@11

"So it is hard to code an algorithm that can determine whether a position is drawn by the mere impossibility of checkmate"
++ Artificial intelligence can nowadays much better diagnose if a skin patch is or is not cancer than any medical doctor having studied for more than 10 years.
So artificial intelligence should be able to see if a position is or is not a dead position better than a mediocre arbiter can do at a glance.
Just train an artificial intelligence.

So you'd have to specify that if one of the bishops is at or below the opponents respective locked pawns rank, it's only dead if the other side has only 1 bishop. But if the opponents bishop is also passed the other sides rank with the locked pawns, or one of multiple bishops is, then it is dead again. Then you also have different configurations of locked pawns that can cause dead positions, but in these you couldn't have bishops included:

So you see it is very hard to program a comprehensive code that included all possible dead positions.

@13

"it is very hard to program a comprehensive code"
++ But it should not be hard to train an artificial intelligence with a set of known dead or nondead positions.

Chess programs are different. Medical research programs other scientific fields just require fast number crunching and handling a million variables, but chess requires actual logic and understanding of positions often with subtle and obscure patterns/strategies. Stockfish supposedly gets "trained" as it plays games and has something similar to a neural network, yet it can't solve this simple forced mate:

We have threads on this topic too, "mates that are difficult for engines" and "too difficult for computers". While positions are obviously easier to determine if they are dead drawn or not than calculating forced mates, there are still so many parameters needed to check. Even with just kings stuck behind locked pawns, with no other pieces, there's so many different configurations of locked pawns, the pawns can be on different ranks, they can be a slanted line like above. Whether bishops can be added or not depends on the specific pawn configuration, what colors the bishops are on, how many bishops, and whether some or all are on specific sides of the locked pawns. It would be an interesting challenge for someone to write a program for it!

@15

"chess requires actual logic and understanding of positions often with subtle and obscure patterns/strategies"
++ Do not think in terms of algorithm. Think in terms of pattern recognition.
When artificial intelligence is fed images of skin patches that are or are not cancer, then after training it can diagnose itself if a patch is or is not cancer better than a medical doctor.
Feed it dead positions and non dead positions and it will be able to tell dead position or not better than an arbiter.

Your assertion would be more plausible if you did what you suggest. You say it should not be hard.

How would you plan to test if it works infallibly? (We already have examples of less than comprehensive code.)

Are you sure about that? I tried playing the computer (in learn->analysis) from this position and it let me complete the mate. Try clicking the magnifying glass then the "Practice vs Computer" button.

It's live chess that doesn't implement it correctly. Analysis even allows the set up of illegal positions so it's not relevant here.

Pattern recognition might work provided you program all the parameters correctly, and any interaction effects as well, such as one parameter affecting whether another still holds true, and then a 3rd parameter affecting WHETHER the 2nd one affects the 1st one..etc. For example, some could be programmed easily:

1. Locked pawns on alternating files same rank.

2. Kings at or below their respective 3rd ranks. (A broader algorithm could be locked pawns on any rank and the kings below the respective pawn ranks instead of "4th rank".) This accounts for the basic locked pawn pattern on any rank.

3. Slanted locked pawns patterns where thr controlled squares controlled by both sets if pawns form two straight lines that the king cannot cross (coordinate algorithms could account for this).

4. Bishops on same color as locked pawns square if parameter 1 holds true (slanted pawn lines in dead positions cant also have bishops).

5. If multiple bishops of the same color, both bishops can't be below their respective 4th rank while the opposing bishop is opposite color and also below it.

This might cover 10% of dead positions. Now you gotta include doubled pawns, knights stuck in the corner behind locked pawns, positions with both bishops and knights stuck (like my complicated one I posted above). Positions where pawns can move but will still end up stuck...we're talking tens of thousands or maybe even over a million possible dead positions, but the basic ones could be accounted for for sure.

To correctly program insufficient mating material, one can do elimination. There are only a few piece combinations where mate is impossible for one side.

KNKQ

KBKQ

KBKR

These are the only 3 combinations where white can never mate in any position. So if black runs out of time, it's a draw. Literally anything else that's not a dead position or inevitable stalemate/self-mate, is a win for white. A King and knight vs a king and 9 queens and 2 rooks is a win for white if black loses on time. But a white king and white bishop vs a black king and black rook, that's a draw if black loses on time.