Assorted tweaks

[Lyudmil Tsvetkov]

Another decent attempt, maybe even much better than the above suggested one, is to give the bonus in case a player has at least one pawn on a5,b5,c5,d5,a6,b6,c6,d6, and at least one pawn on h5,g5,f5,e5,h6,g6,f6,e6.

That is, when you have at least one pawn on one of the 8 squares on the 5th and 6th ranks on the queen side, and at least one pawn on one of the 8 squares on the 5th and 6th ranks on the king side, you give the bonus.

[d]6k1/1p3pp1/1P6/5P2/8/8/8/6K1 w - - 0 1
for example, you give the bonus here, as white has one pawn on b6 on the queen side and one pawn on f5 on the king side

If all those squares are considered, I think the bonus could go a bit up, maybe 15-20cps normal values instead of just 10, due only for the mg.

I think this is definitely the better and right way to apply space advantage on both sides of the board - a very reasonable term if implemented in this way, and I absolutely do not see what objections could be raised to such an approach. It is as scientific as penalising an isolated pawn.

[Lyudmil Tsvetkov]

Any bishop in the mg in front of own pawn on e2 or d2 is due some reasonable penalty, maybe 5-10cps. It is obvious that such bishops thwart severely the development of the own center, which is bad.

Of course, the penalty will be due just for the mg.

However, this is the only case when minor or other pieces in front of own pawns on the 2nd rank are due penalty in the mg, for example any minor in front of own pawn on a2,b2,c2,f2,g2 or h2 looks quite normally. Knights on e3 and d3 with pawns behind on e2 or d2 also are much more acceptable than bishops on those squares. So this rule should be specialcased. A bishop on e3 or d3 with such a constellation is really an odd-looking enterprise.

[d]6k1/8/8/8/8/3BB3/3PP3/6K1 w - - 0 1
penalty is due for both Be3 and Bd3, blocking own central e2 and d2 pawns

Btw., engines quite frequently go wrong with this rule, even the top.

[d]r1bqk1nr/pppp1ppp/2nb4/1B2p3/4P3/5N2/PPPP1PPP/RNBQK2R w KQkq - 0 1

Anyone having seen an engine playing the Ruy Lopez like this?

[Lyudmil Tsvetkov]

Here my vision how minor outposts should be scored.

- files b through g for outposts on ranks 3,4 and 7
- files a through h for outposts on ranks 5 and 6

[d]6k1/7p/p5pB/Np6/1P6/8/8/6K1 w - - 0 1
I would score Na5 and Bh6 as outposts, but not if the minors were on a4,a3,h4,h3, or a7,h7.

Of course, minor outposts on the edge files would be due significantly lower bonus than for g and b files, maybe even twice lower, but not giving any bonus at all apart from psqt for minors on the 5th and 6th ranks is obviously also not the right decision, as frequently, although edge, such minors are rather strong.

At the end of the day, it will all boil down to tuning outposts, psqt and different other parameters, but having an acceptable vision before starting tuning is maybe the slightly better approach.

I wonder how many engines score edge outposts?

[Lyudmil Tsvetkov]

We talked about that, but, as it is very important, and one of the things top engines fully do not understand, briefly again.

I call a phalanx pawn an intermediate/connecting pawn of a larger group of pawns, consisting of at least 3 pawns. There are different definitions for what a phalanx pawn actually is, but, for me, you can not have a phalanx pawn in a group of just 2 pawns, in the same way as you can not have a phalanx of soldiers consisting of only 2 soldiers - the smallest phalanx of soldiers consists of at least 3 soldiers, no matter what their arrangement is.

So, below I will use the word phalanx pawn for intermediate pawns of larger groups of at least 3 pawns, but, if anyone decides to bring me to court over the issue, for example Pawel, then I will gladly use simply the word larger group pawn.

Once again the definition for a phalanx pawn: a pawn that has 2 own pawns on adjacent files, each one of which is on the same rank, one rank behind or one rank in front, so adjacently placed, in touch with the phalanx pawn, is called a phalanx pawn.

[d]6k1/5pp1/1p6/p1pp4/5PPP/3P4/1PP5/6K1 w - - 0 1
c2 is a phalanx pawn above
g4 is a phalanx pawn above
b6 and c5 for black are phalanx pawns

again, as you see, all those pawns have 2 adjacent own pawns

Please read this twice, and then a third time - 100 elo to gain from that.

[Henk]

Any bishop in the mg in front of own pawn on e2 or d2 is due some reasonable penalty, maybe 5-10cps. It is obvious that such bishops thwart severely the development of the own center, which is bad.

Of course, the penalty will be due just for the mg.

However, this is the only case when minor or other pieces in front of own pawns on the 2nd rank are due penalty in the mg, for example any minor in front of own pawn on a2,b2,c2,f2,g2 or h2 looks quite normally. Knights on e3 and d3 with pawns behind on e2 or d2 also are much more acceptable than bishops on those squares. So this rule should be specialcased. A bishop on e3 or d3 with such a constellation is really an odd-looking enterprise.

[d]6k1/8/8/8/8/3BB3/3PP3/6K1 w - - 0 1
penalty is due for both Be3 and Bd3, blocking own central e2 and d2 pawns

Btw., engines quite frequently go wrong with this rule, even the top.

[d]r1bqk1nr/pppp1ppp/2nb4/1B2p3/4P3/5N2/PPPP1PPP/RNBQK2R w KQkq - 0 1

Anyone having seen an engine playing the Ruy Lopez like this?

Yes Skipper likes to play Bd6 or Be6 when its pawns are on d7 or e7. And it never plays grunfeld.

[Lyudmil Tsvetkov]

There are good, neutral and bad phalanx pawns.

It is especially important to score good phalanx pawns.

Again, a duo pawn, a pawn adjacent to other own pawn on the same rank, and a defended pawn, are the primary good elements of small, 2-pawn groups.

For phalanx pawns, intermediate pawns of groups of at least 3 pawns, those same elements will be covered and receive overbonus when they occur in such pawns twice, so not to have loose links.

[d]6k1/p6p/1p3pp1/2p5/8/1P2PPP1/P1P5/6K1 w - - 0 1

above, b3 is a good phalanx pawn, two times defended
f3 is a good phalanx pawn, 2 times duo
g6 is a good phalanx pawn, one time duo pawn and one time defended pawn
b6 is a good phalanx pawn, one time defended and one time defending, but, as it defends another defended pawn, we might consider it for at least 2 times defended

All of the above pawns deserve some nice overbonus besides the already given points for duo pawns and defended pawns, as they link together other own pawns in a convenient way.

The bonus should be preferably given in terms of files and ranks.
There are specificities about each of the above pawns, that I have elaborated on in other threads, they are not all equally strong.

[d]6k1/8/8/ppPPPPpp/8/8/8/6K1 w - - 0 1
for those who do not believe in larger groups of pawns and phalanx pawns, please look at the above diagram, but look very carefully.

I placed the pawns so to be on the same rank, so that you can quickly spot the difference; the fact that the white pawns are a bit more advanced is not what you should be looking at: imagine they are on precisely the same rank in terms of advancement.

4 pawns for each side.
Do you think they are equally strong?

Of course not, the white cohesive group of 4 pawns all adjacent to each other is much much stronger than the 2 separate small black groups of 2 pawns each. That is an irrefutable chess knowledge reality.

And indeed, when you consider phalanx pawns, the white group has 2 such pawns, d5 and e5, while the 2 small black groups lack even a single phalanx pawn. When you give the phalanx bonus for those 2 pawns, you will know that white is much better.

Same on the battlefield - everyone knows that a single bigger unit of forces that cooperate with each other is far superior to separate smaller detachments, although they could have the same overall personnel and equipment, but lack the ability to cooperate with each other.

[Lyudmil Tsvetkov]

Besides from giving bonus for good phalanx pawns, and possibly malus for bad phalanx pawns like double bases, you should also, if you want to be correct on evaluation of larger groups of pawns, consisting not only of 3 pawns, but 4, 5, 6 pawns and even bigger, consider some small bonus, maybe some 7-8cps in the case of any phalanx pawn at all, no matter if it is good, neutral or bad. Why so?

Well, the good pawns will already get their bonus, the bad ones possibly too their deserved penalty, but, apart from that, you need to score any phalanx pawn at all, as, in larger groups of pawns consisting of 4,5 and more pawns good and bad pawns will alternate frequently, but that will not change the fact that the group is still very large. Very large groups need their separate bonus, as they are better on average than smaller groups of pawns, although the respective phalanx pawns for each group might exhibit similar or even exactly the same characteristics. You simply need to reflect the economies of scale. Any phalanx pawn bonus does precisely that.

[d]6k1/5p1p/6p1/ppp5/4P1P1/2PP1P2/1P6/6K1 w - - 0 1
above, the white group of 6 pawns is much superior to the 2 black groups of 3 pawns each.

in the white groups, you have 4 phalanx pawns in all: c3,d3,e4 and f3; in the 2 black groups you have just 2 phalanx pawns - b5 and g6

It does not matter at that, that black's phalanx pawns are both good, while for white you have 2 good - c3 and e4, one neutral - d3, and one bad - f3, a double base. The white groups is still superior because of its abilities to cooperate between the pawns. Of course, it would be even better if there were not bad pawns in the larger group. Each element receives its penalties and bonus points - the any phalanx pawn bonus is given precisely for the fact that it recognises a very large group of pawns.

A very large group of pawns of course moves forward, cooperates and restricts the enemy pieces' activity much more efficiently than smaller groups of pawns.

[d]6k1/1pp2pp1/p3p3/6P1/2PpPP1P/P2P4/8/6K1 w - - 0 1
white is immeasurably better above in terms of pawn eval, even if ranks are not considered

some engines might think, well, black does not have an isolated pawn, just d4 might be scored for vertically isolated/derived from the group, while white has a3 which is isolated, so let us give the penalty to a3, call eval and show the score in black's favour.

Thousand times no - as said, white is immeasurably better in terms of pawn eval.
Above, white has 4 phalanx pawns, d3,e4,f4,g5, while black just 2, b7 and f7. The white groups of pawns, although lacking an isolated pawn, is much more cohesive, advances more efficiently and restrcits the enemy pieces' activity much better.

Such large and very large cohesive groups of pawns are also especially suited for attacking the enemy king, as you can see above, where the black king is really in danger. There is nothing more efficient and irresistible than a very large cohesive group of pawns attacking the enemy king. Unfortunately, quite often, when attack is done in this way, the associated lines are deep or very deep, as the group advances slowly but steadily, in distinction to just quickly pushing some pawns to open files, so engines simply fail to understand the worth of such terms.

Phalanx pawns, large cohesive groups of pawns, are one of the main weaknesses of current top engines, and, if this is true of the top, it is true of all engines.

If I were an ardent programmer, I would read this twice and a third time, because it is very important.

Again, for such terms you need much more search depth, as the associated lines are deep or very deep in distinction to quick pawn pushes opening lines, where engines see everything, maybe testers should test at 5 minutes instead of seconds, but, what is beyond any doubt, is that the abovementioned terms are indeed very useful, if you also have the luck to tune them properly.

[Lyudmil Tsvetkov]

Very simple and very true chess rule: in the late endgame, when there is a lone queen vs enemy rook + minor, scale down the score.

Q vs R+B - scale down score by 50%
Q vs R+N - scale down score by 30%

Quite obviously, most of this endgames will end in a draw, so better leave aside those lines for some other lines, that might actually be more promising.

[d]6k1/1p1r3p/p1b3p1/8/8/P3Q3/1P4PP/6K1 w - - 0 1
draw

[d]6k1/5p2/1p1r2p1/p4n1p/7P/P5P1/1P2QP2/6K1 w - - 0 1
considerable drawing chances

Some might think that this rule is stupid, but that is not so - actually it might prove very useful, as such configurations are much more frequent than Q vs 3 minors, for example.

[Lyudmil Tsvetkov]

Whenever you find an inner pawn of a long chain, whose defending pawn and the pawn that it defends are both blocked by enemy pawns, you might consider this pawn as backward and give some 10cps penalty at least in all cases.

[d]6k1/8/5p2/4pP2/3p4/3P4/8/6K1 w - - 0 1

you might consider e5 as backward

obviously, it can not move forward without being lost, and, although it is defended by f6 and has own pawns on less advanced ranks on adjacent file, this pawn, f6, is blocked and can not realistically support the advance of e5.

I think this is an useful rule for very advanced engines, but how you tune it to all other terms fully transcends my imagination.

[Lyudmil Tsvetkov]

Try this rule of splitting bishop pair evaluation between opposed bishop pairs - the case when the enemy has a bishop, and unopposed bishop pairs - the case when the enemy does not have a bishop.

An unopposed bishop pair is simply much stronger, as an enemy bishop kind of minimises the efficiency of the pair.

If you have 50cps bonus for any bishop pair, an opposed bishop pair might get 40cps, while an unopposed bishop pair 60cps.

I think this rule would be true in most cases, however, bearing in mind that you have to tune it to piece values and other existing imbalances...