Mephisto 3.1.209 — puzzle pacing and ratings
Puzzle Mode: pacing, a replay guard, and the puzzle's rating on screen.
Which download?
| Archive | Size | Use it when |
|---|---|---|
mephisto-3.1.209.zip |
585 MB | First install — everything, engines included |
mephisto-3.1.209-update.zip |
5.9 MB | You already have Mephisto — extract over your existing folder |
⚠️ Extract the update over your existing install, never into an empty folder — without the engines it cannot run. Extract in place: Chrome derives an unpacked extension's id from its folder path, so a new folder means a new id and re-running the native-host installer.
Two ways a solved puzzle was being thrown away
Playing twice from the same position. After your move the puzzle scripts the opponent's reply, and until that lands the board still shows the position you just moved from. A second move issued into it answers a question that has already been answered — on a graded puzzle that is a wrong answer, not a no-op. A move is now refused if it would be played from the position we last played from, in a window deliberately shorter than the 1200 ms retry, so a click that genuinely failed can still be re-issued.
No pause on the engine path. The database path already waited; a searched puzzle move went out the instant the search returned. Every puzzle move now waits 300 ms, Lichess and Chess.com alike — including the case that matters most, where the database has no answer and the engine has to supply one.
The puzzle's rating, beside the move
Puzzle solution: a4b4 (Rating 2145). Both publishers provide one; daily archive rows do not and simply show none. Records written before this release are still read correctly — they just have no rating until that database is imported again.
One caught before it shipped
The pause and the replay guard first sat in the same block, so the delayed continuation re-ran the guard, found the record it had written 300 ms earlier, and dropped the move. That would have killed every engine-played puzzle move while leaving database moves working — the hardest kind of bug to notice. The guard now runs only on the way in, and the test suite drives it for real rather than grepping for it: the source-text checks proved the guard existed, not that a move survived it.