Bug in Explorer - Impossible Move

Open the Explorer, and make the moves:

1. e4 c6 2. Nf3 d5 3. d4 dxe4 4. Ng5

https://www.chess.com/explorer?moveList=e4+c6+Nf3+d5+d4+dxe4+Ng5&ply=7&origMoves=e4+c6+Nf3+d5+d4+dxe4+Ng5

Here the explorer lists that two master games have the move 4 ... exd3

Of course, this is illegal, and the explorer won't let you make that move. If you open up the two master games you will see that the position is different, and there is a pawn on d3 in both games.

I reported this as a bug, and was given a stock answer saying it was not a bug and that the impossible pawn move was due to en passant, which is clearly wrong. The last move was a knight move (Ng5), and the pawn on e4 wasn't even there when the d4 pawn moved. En passant does not apply here.

Anyone else care to test and confirm?

It's possible in other move order:

Thanks for testing.

Yes, it could be reached in another move order but the explorer is, I think, only supposed to suggest next moves that are legal from the existing position? Normally one can click on the suggested move and the explorer will make the move - in this case it does nothing because the move exd3 is illegal.

It's also rather odd/broken that the two games listed for exd3 are for a different position where d3 was played and not d4. That's not a different move order, but a completely different move.

Here's another example. Both positions look identical, but are reached by a different move order. In the first, en passant is possible, in the second, it is not.

1. https://www.chess.com/explorer?moveList=e4+e6+e5+d5&ply=4

2. https://www.chess.com/explorer?moveList=e4+d5+e5+e6&ply=4

They even show different names in the Explorer, so it knows they are different positions.

1 is called the "French Defense: Steinitz Attack"

2. is a Scandanavian.

This isn't a transposition, because different moves are possible.

In both cases Explorer shows exd6 as a possible move from Master games.

Pick one of the 42 games (click on the number 42) to get the list:

https://www.chess.com/games/search?fen=rnbqkbnr/ppp2ppp/3Pp3/8/8/8/PPPP1PPP/RNBQKBNR%20b%20KQkq

and look at the moves in the game. It's not just a different move order, but the games show *neither* position 1 nor 2, but a completely different third position where the pawn is on d6.

Is that really the expected behaviour? Seems like a clear violation of the Law of Least Astonishment :-)

Explorer is not move dependent, but position dependent. So transpositions will show up like that, where you'll see a position that will match the current position but one ply different, regardless of what move was made to get there.

Pretty sure most databases work in a similar way.

Thanks for your response.

OK. Even if we accept that transpositions will show up, and therefore it is going to show impossible moves for the position (en passant where it is illegal), that still doesn't address the bug that Explorer is showing games for a different (non-transposed) position.

In the examples I gave, the Explorer will list the games where the suspect move was played, and if you look at those games, the position is different. ie. the pawn is on a different square. That's not a transposition - it's a different position entirely.

eg.

1. e4 d5 2. e5 e6

Here, the pawn is on d5. Ignoring the move order, and the fact that exd6 is illegal, Explorer is still showing incorrect information.

if you click the "42" next to the move exd6, it will take you to a list of games where that move (supposedly) was played from that position. Except it doesn't. It shows this list:

https://www.chess.com/games/search?fen=rnbqkbnr/ppp2ppp/3Pp3/8/8/8/PPPP1PPP/RNBQKBNR%20b%20KQkq

Click on the first game. Here the pawn is on d6

1. e4 e6  2. e5 d6 (not d5)

So, of course here exd6 is possible. It's a different position.

The Explorer should only show legal moves from the position. If a king moved, then moved back, it shouldn't show castling as possible from that position. That's just broken. FIDE rules consider these positions to be entirely different eg. for purposes for repeating positions ending in draws.

Explorer clearly can tell the difference between the positions as it names them differently.

If the database is purely positional, and move order doesn't matter, and should therefore show illegal moves, lets consider the immortal Magnus vs Hikaru "Bongcloud" game:

https://www.chess.com/games/view/16025283

Make the moves in the Explorer:

1. e4 e5 2. Ke2 Ke7 3. Ke1 Ke8

https://www.chess.com/explorer?moveList=e4+e5+Ke2+Ke7+Ke1+Ke8+Nf3&ply=6

Because it's tracking the move order, Explorer shows this as "King's Pawn Opening: The Bongcloud", not "Kings Pawn Opening".

But we'll ignore that and play 4. Nf3

Explorer doesn't ignore the transposition here - it says there are no games found in the database with this position and suggests no moves at all. Apparently no master has ever played an Italian!

If consistent it would be showing all possible moves from this position. ie. all games from 1. e4 e5.

We can't have it both ways. Explorer needs to be consistent. Either it shows illegal moves and games that will (but haven't yet) transposed (because it's positional only), or it should show only moves that are possible from the position as reached by the move order (a lot more useful imho). What it does at the moment is rather confusingly somewhere in between.

In that position, they aren't the same. Castling rights are part of the criteria if a position is the same or not as another position. In the bongcloud, castling isn't legal, so it will never show positions where castling still is legal, even if the pieces go back to their starting squares.

OK. So different castling rights == different position. Got it. That's what I thought. So why would different en passant rights not also be a different position?

FIDE 9.2 says "and the possible moves of all the pieces of both players are the same".

That seems to imply that either a difference in castling or en passant rights equals a different position.  ie. a difference in either should be taken into account by the Explorer and only legal moves should be shown.

At the moment, a difference in castling rights is taken into account by Explorer (and only legal moves are displayed), but en passant rights are ignored (and thus illegal moves are displayed). That is inconsistent.

Ultimately it's the resultant position that matters for Explorer. So it's possible that en passant isn't possible in the first position, due to the last move not being a pawn move, but one ply deeper, that resulting position did have a pawn move that allowed it: as above the difference between d4 having already been made and the positions where the the change in position was reached by d4 allowing en passant.

I don't know how explorer handles the move order in such cases, but the resultant position should recognize the FEN and en passant wouldn't be possible at that point.

I know there has been at least one bug related to this that has been reported, but generally, it should just be position based.

Agreed. It should be position based (noting the previous point about positions only being equal where movement rights are the same).

Utlimately I use Explorer to answer the question "from this position (however reached), what moves have happened in master (or my own) games and how often?" Also it is useful to pull up the list of games from that position to dig in further.

For the explorer to include moves that aren't legal or which rely on a different physical position of the pieces is confusing at best. For it to include illegal moves for en passant, but only legal ones for castling is doubly so.

It's a great tool, but has a few little glitches. eg. switching from "master games" to "my games" or vice-versa confuses it (but clicking back then forward sorts it out).

I use Explorer and the analysis tools a lot. I find it facinating to see how how one opening transposes to another. I had a game the other day transpose from a Caro-Kann Panov into a Grünfeld variant which surprised me, but it appears to be pretty easy to reach one from the other.