reducing too much

[hgm]

I guess this can indeed happen, if you don't play the expected (i.e. PV) move (irrespective of reductions). In this other branch, the engine must also have a mate against you, perhaps even a faster one, but the score in the hash will only be a lower bound, as it has not yet exhaustively searched if because of your deviation, there is not one that is even faster than that. So at low depth it cannot use the hash score, starts re-searching, an if there was no much faster mate, initially does not find a mate.

It would only be cause for worry if it announced a mate-in-N within the horizon, and on the next move, once you let it deepen enough to see mate-in-(N-2), would not find this mate. If the initial mate was over the horizon (seen by transposition-table grafting), then there is no guarantee it will find it again. Such mates are very illusive, and happen on a lucky search order of sub-optimal moves, and the survival of their entries in the hash table,

[AndrewShort]

...It would only be cause for worry if it announced a mate-in-N within the horizon, and on the next move, once you let it deepen enough to see mate-in-(N-2), would not find this mate. If the initial mate was over the horizon (seen by transposition-table grafting), then there is no guarantee it will find it again. Such mates are very illusive, and happen on a lucky search order of sub-optimal moves, and the survival of their entries in the hash table,

Just to be clear, are you saying if an engine completes a 12 ply search, showing forced mate on ply 11, then on the next move, it completes a 10 ply search, that the engine is _guaranteed_ to find the continuation no later than ply 9? (I ask, because reductions mean a mate in the horizon can be missed) or might it need a 12 ply search again to continue the forced line, due to reductions?