Well, as I already told you, I am unable to think properly right now. However, I wanted to post a probably last small portion of evaluation topics, 5 or 6, because there are definite chances that in the near future I will be busier and have even less opportunity to do that.
The topics have not been treated here, I think they are quite original and important, so I would really like to share them with you. Most of the stuff was already present on the pages of the compendium I wrote, but they were so poorly explained and with clumsy implementation suggestions, that I think no one paid attention to them. Here I hope to give a bit clearer interpretation.
I will apologise again for any inaccuracies in the posts. It is all due to the noise. As I will be lacking the necessary concentration, I am thinking of just sketching them, maybe someone will find something useful even that way. With some luck, I might even be able to develop some of the topics. Of course, if Arjun is not still around.
As I told you, I will be sketchy. I hope you grasp it.
The concept of an outnumbered pawn (do not care about the name, I have to call it somehow) will be a pawn that has 2 (at least 2, there might be 3 or more) enemy pawns immediately adjacent. Immediately adjacent will mean on one of the 8 immediately adjacent squares as attacked by the king. The logic behind it is that such pawns are usually more vulnerable, because the enemy pawns nearby restrict their solidity and value.
[d]6k1/8/8/P1p2p2/p1Pp1PpP/P5P1/8/6K1 w - - 0 58
Above, a4 will be outnumbered (a3 and a5 immediately adjacent), white c4 also, c5 and d4 immediately adjacent, and also white f4, f5 and g4 adjacent, and black g4, g3,h4 and f4 adjacent.
An outnumbered pawn will of course, wherever found, be penalised. I would say some 5-15cps penalty would be fine (depending on what other related terms you have).
Again, do not care about the name. A deficient pawn will be a pawn that controls/attacks the squares in front of 2 enemt pawns.
[d]6k1/2p1p3/3p4/3P1ppp/8/6P1/7P/6K1 w - - 0 58
d5 and g3 above are deficient pawns; d5 controls c6 and e6 squares in front of c7 and e7 respectively, while g3 similarly controls the f4 and h4 squares in front of f5 and h5 respectively.
Such a pawn will receive a penalty of course, some 5-15cps depending on what other related terms you have. The reason for the penalty is that the closeness of more than 1 enemy pawns supposes a portion of weakness about such a pawn, that could be exploited.
Take the white d5 pawn, for example. You might even think that it is very strong because it makes 2 enemy pawns backward, but that is not so. Actually, quite the opposite. Any time soon black can play either c6 or e6, after which it becomes evident that d5 is more of a weakness, it is bound to perish and thus boost the importance of the remaining 2 black connected pawns. Taking this into consideration might help you not overestimate the strength of a pawn making backward 2 enemy pawns at the same time, etc. Similarly with the white pawn on g3.
What do you think of this concept? Even stupider than an outnumbered pawn?
Well, here the penalty is huge of course, but it is quite probable that some other related terms cover that for you. In any case, an immobile pawn on the second rank will be a pawn that lacks any mobility (i.e. can not advance and capture), because it is blocked by an enemy pawn or piece.
Of course, such a pawn is very bad, because it is bound to stay back there for a while and prevent own pieces from developing. I would exclude from the penalty immobile pawns blocked by own pawn or pieces, as in this case it is easier to remove them and free the pawn to advance.
[d]6k1/p2p2p1/p2P2Np/7P/8/8/8/6K1 w - - 0 58
g7 and d7 are immobile pawns on the 2nd rank. a7 also is, but here the blocker is an own object, so I would not consider it.
The penalty for d7 and g7 could be huge, in the middlegame of course, some 20-50cps.
Please note that the presence of other blockers is also relevant, even if they are heavy pieces.
[d]6k1/p2p2p1/B2Q2Rp/7P/8/8/8/6K1 w - - 0 58
g7, d7 and a7 above are all due penalties.
What do you think of this?
Having a better implementation?
Some think doubled pawns are useless. Actually, almost everybody on this forum says so. It is not like that. As the most prominent and easily observable pawn features doubled pawns are bound to be extremely important. It is only that most people probably do not quite know how to use them. Below a couple of remarks that might be not that stupid.
1. doubled and isolated
Such pawns are very unwelcome. I would say, increase the penalty for doubled in this case.
2. doubled and passers
I would neither increase, not decrease the penalty for doubleness here, just consider it. The pawn is a passer, but it is doubled.
3. doubled and candidate passer
I would decrease the penalty for doubled here, as chances are very soon 2 instead of only one passers, even if doubled, might appear. For the time being the engine will see only a single candidate, and that might deceive it. I think this is important, but not sure how many engines consider it.
4. doubled and lever
Of course, here the penalty for doubleness should be decreased, as levers are there to get rid of the doubling. The search might see it, but you never know how deep the search goes, it might terminate just with the lever.
[d]6k1/8/3p2p1/5P2/P1PP2P1/2P2P2/P7/6K1 w - - 0 58
a2 and a4 above are both doubled and isolated. As passers they are normally good, but not so when they are just doubled and isolated and no passers. Then doubleness is felt much stronger.
c4 is doubled and candidate. Currently, the engine will see c4 being only candidate, but actually this situation is bound to translate into 2 full passers sooner or later. For example, after c4-c5 black can capture on c5 and 2, although doubled and isolated passers are created; play d5, and after c3-c4 again 2 passers for white are created; otherwise, white captures on d6. Thus, the power of a doubled candidate passer could usually be underestimated.
f5 is doubled and lever. As f5 would usually be protected also by a white piece, such a situation offers good chances for undoubling. On g6-g5 will follow f3-f4 with another lever and undoubling. Therefore doubled levers are good.
I think you could increase and decrease accordingly the doubleness weight in the different cases to capture real things better, by 20-50%, etc.
Any comments on this? Again, all on this forum are completely convinced doubled pawns are fully useless?
The more I think, the more I am convinced that immobile pawns are much more important than considering mobility for pawns, but the problem is how exactly to do them and to tune the parameters in the larger pool.
There are things that you can not explain sufficiently in eval, if you do not take into consideration immobile pawns.
Hi lyudmil
I find your posts always very interresting .I develop again a new chess engine and mainly I like to play with eval function .I can find many ideas in your posts and in your pdf copendium (?)
Thanks for your work !
Best
Dany
As I will be lacking the necessary concentration, I am thinking of just sketching them, maybe someone will find something useful even that way. With some luck, I might even be able to develop some of the topics. Of course, if Arjun is not still around. 
.I can find many ideas in your posts and in your pdf copendium (?)