My answer below, doesn't take in to account your knowledge of the parameters you have at your disposal to adjust {statically} so if we can agree for a moment that for each parameter you have a static value set, to achieve the best balanced tuning across many of the tests you carry out, we have instead a data range of values that can be dynamically applied to each of the same parameters, based on a set of input value combinations; effectively changing the behaviour of Komodo by whatever amount you decide to be sufficient for the particular effect each parameter has - make sense so far?

So, to name a few 'Inputs' we have:
(1) Opponent history (where Komodo knows this and already has a 'tuned' style against this opponent)
(2) Opponent move choices (error rates, i.e., how do moves played compare to Komodo's 'best move' table - how large are these errors)
(3) Opening played
(4) Position assessment (complexity, open/closed, etc)

My thinking on this is that by taking in to account a number of inputs that in turn can affect the playing style of Komodo, you end up tailoring the performance to what is in front of you rather than the 'best fit' approach that may not be too good in some cases {e.g., imagine from the input Komodo is assessing the position to be closed then adjusting the parameter or parameters to make Komodo play less tactically with more emphasis on positionally play, such as, nice squares for pieces}

NB: I guess (2) above is my attempt at trying to come up with a way Komodo could check what I and I imagine you do OTB, i.e., look to see what my opponent has played, primarily to identify and exploit any drawbacks that I can see from the move played.

Any use?

Probably the easiest advantage to exploit would be having a significant clock advantage over the opponent. In this case, one could attempt to reduce opponent's ponder hit rate (when nearly equal move choices were available) and give preference to lines that have a significant score shift between the opponent's expected time per move and own time per move.

They are two different concepts. I think we already do enough (or almost enough) about the king, but the initiatve is not really a concept that is in any program to my knowledge. Maybe it should be.

I think there are three aspects to aggressiveness:

1- King Aggressiveness: Aggressiveness towards the enemy king, what pieces attack its surrounding squares, what pieces can check it or make it play forced moves (e.g. "play the king from c8 to a8 or die"), how close are they to it, etc. I think this setting should be asynchronous, for white and black, just because you want to attack the opponent's king it doesn't mean you're obsessed with defending yours. Zappa Mexico has this well covered, but plays poorly on the defensive, so this parameter is only useful when you are on the attack.

2- Activity Aggressiveness: About having the initiative, mobility, and restricting the mobility of the opponent. Getting passed pawns may be included in this, as without having attacks against the king another way to win is to end with a passed pawn with your initiative and convert to a win. I think Rybka 4 Mindbreaker's setting is excellent for this, without drawbacks as you can set one for the black or the white side, though of course you have to know when being more passive or patient may be better.

3- Complex Aggressiveness: In where the goal is complicating the position and defeating an opponent in a position that you play better. This includes pawn sacrifices, piece sacrifices against pawns, sacrificing the exchange or forcing the opponent to sacrifice theirs, Queen for 2 rooks or 3 pieces exchanges, etc. Material imbalances get huge bonuses for the sake of imbalance, so if you see a variation where you end three pawns against piece that is 0.20 worse than something else, you go for it anyway. The catch is that if the opponents avoids going for that position he ends worse than if you didn't go for it. It's "aggressive" because it decreases the chances of a drawn game significantly. Rybka 3 Dynamic at high contempt is a great tool for seeking these kind of positions, but against a stronger opponent this kind of playing is bad because you may end up being the loser.

Kind 2 and Kind 3 may run against each other, while Kind 2 keeps its own pieces on the board (there can't be high mobility if the pieces are exchanged) and probably seek closed positions that maximize the power of individual pieces, Kind 3 looks for open positions that lead to the material imbalances.

> Please could you help me by proposing an engine you regard as fairly strong (>2700 elo) but also fairly passive in its playing style. I will carry out some tests on this engine and I'll see if the idea works.

(Personal suggestions: Rybka 2.3.2a and Naum 3 - those wee regarded as in the limit of highest strength and highest passiveness.)

If aggression includes willingness to sacrifice material for less tangible factors like attacking chances against the king, you could not get a cleaner example than by comparing Komodo 2.xx (say 2.03) to Komodo 3.0. The change in this aspect was quite dramatic.

For analysis, most people allow some number of minutes per move. Therefore the best measure of analysis strength is long time limit tests, and preferably ones where the time per move is fairly constant, removing the importance of time control optimization. This should be clear to everyone. Unfortunately no one tests at anything like 1 minute plus 3 minutes per move increment. The closest we have is 40/2 hour results, with 40/40' being next best. For those who want quick answers or to run their own blitz matches, a large increment relative to the base time will better simulate analysis. For those who want to run their engines against others on Playchess or elsewhere at blitz, the IPON list is probably the most reliable rating.

from what i hear from Larry,H2 is weaker at long time control (at that list) than H1.5a