When an engine says +1.00 is that the equivalent of a pawn?

LikeTheLake - the point is that the notion of "exactly one pawn" is semantically meaningless because you have no positional critieria by where you can say "taking away a pawn here" or "adding a pawn here" gives us exactly a value of 1.0 that is consistent across engines. You have no ostensive definition of what the value of a pawn is. So your statement comes down to "1.0 means whatever the chess programmers for that engine decides it means." 

Unless and until you can show me how to find the "exact value of a pawn" on an engine, your statement is semantically meaningless. 

Well that's the problem right?  The only real evaluations are white is winning, black is winning, or draw.  Even in my personal evaluation I don't just count up the pieces.

It's better to say 1.00 is the equivalent of a pawn.  If we want to get really picky we can note a pawn isn't worth anything itself.  The position is either winning or drawn.

Which is what the Houdini team did -- as you note, they tuned their engine evaluation such that +1.00, +2.00 and +3.00 relate specifically to winning percentages NOT to material advantages (even though they still refer to the numerical value as "pawns" out of convention).  

I have no problem if someone wants to say "+1.0 means a pawn, give or take" But if someone says it means "EXACTLY the value of 1 pawn" -- as if that value is knowable or meaningful then I'm going to nit pick. Because (a) such a value isn't knowable and (b) it is clear that for some engines that isn't what the 1.0 means. 

It's just confusing to me how they arrived at these percentages... but I guess it's slightly less vague than saying "the equivalent of a pawn" heh.

They played lots and lots of games from given starting positions on identical hardware using identical settings, and then back tuned the evaluation engine such that they got evaluations that matched the results. 

Haha, sometimes it seems like houdini only values a pawn at a few "tenths of a pawn" (like .30-.60) based on my experience. Probably not true of course, but the engine must take a lot of things into account that people perhaps don't give it enough credit for.

I think you make some good points, Kingpatzer. The value of a pawn, as with pretty much any feature of a position, is hardly static.

I simply rate +1.00 as "White has a really good position."

Houdini, more than any other strong engine, would rather have play than material (although all engines still favor material), such as a pawn. That is why in your example with a2 houdini gives white a better evaluation than stockfish - it sees more counterplay down the a file and with the move. Also, stockfish tends to overexagerrate advantages (give higher evaluations) in my experience.

Shepi13 I'm not arguing that engines evaluate anything identically. I'm arguing that they don't, which is part of the reason why "the value of exactly one pawn" is a semantically meaningless statement with respect to engines. 

I would not believe these scores too much unless you really see a combination that results in a real material difference no farther away than 3 moves or so. In my club we found that 4 guys together can make these scores tip in a couple of moves. I do not know how the weighing works, but we found scores between -1 and +1 often difficult to support.

Once again, the base unit in chess is a pawn. So 1= 1 pawn.

But computers even give a different value to the pawn depending on what square it's on. Still, the starting point is the pawn. There are many other factors at play, which is why most engines show evaluations at least two decimal places (1.00).