OpenChess

I have 3-4-5 gaviota TBs. While running Houdini 1.5ax64 in Arena, the following position
8/8/8/3k4/4NK2/5p2/5N2/8 w - - 9 68
is not recognized as "tablebased". This is a bit weird, since all connected positions are properly recognized (as present in tablebase), and gaviota TBs have always worked flawlessly on my PC.

By the way, I have just registered to the forum... the question "are you a spammer?" was quite hilarious.

As a general principal, Houdini deliberately does not use GTBs when it "seems" to have winning chances even though the tablebases see it a draw. It does this because it hopes the other side does not have tablebases and might be tempted into a blunder if Houdini makes a less than perfect move.

It is an annoying waste of time when Houdini is playing another engine because tempting another engine into a blunder is very unlikely. It appears the strategy is for use against a human.

This "strategy" is a Crafty idea that has been around for 15 years now. In version 15.x, "swindle mode" was introduced for that very reason, that when you reach KRP vs KR that is not always easy to draw, a program can't differentiate between KRKR (draw) and KRPKR (draw, but much harder). Disabling EGTB probes, making sure that at the root, the only moves you consider are moves with a draw score so that you don't trick yourself into playing a losing move, lets you choose from the drawing moves, but with a normal search and backed up score, so that you don't just throw the pawn away, and give the opponent a chance to make a mistake.

In Crafty, this can be disabled (swindle=0 in .craftyrc/crafty.rc file) for engine vs engine games, unless your opponent doesn't use EGTBs, then it should be left on (default swindle=1)...