5-1 EGTB generation revisited

[Martin Kreuzer]

Hi Guy, Guido,

attached I post the max DTM positions for KBBBBK which
I extracted from the databases using FEG -LL

For wtm, the DTM seems to be 13 and there are more than 100 positions
(FEG lists only the first 100 or so)
For btm, the DTM seems to be 16 and there are 8 positions.

In the Christmas break, I will make more detailed experiments
(also with other endgames) and post the results here.

Greetings,
Martin Kreuzer

[guyhaw]

Thanks Martin. I'm only looking myself for one position for the spreadsheet but ...
it will be good to see, with the small number of maxDTM positions, that they are indeed actually distinct from each other. The '8' btm KBBBBK positions is what I was expecting (192/24).
However, you have 96 wtm when I was expecting 80 (1920/24). The wK is resolutely on a1 and the bK ditto not on a1-h8 so my usual suspicions about '2 Ks on the diagonal' are not confirmed here.
g

[Martin Kreuzer]

Hello Guy,

there were actually more than 96 positions, but FEG stops listing them after approx. 100. For instance, there was a "partialy printed" position No. 97 which I deleted. I will try to redo the computation and let it run to the end but there will probably not be more than the first 100 positions in the file.

Cordial greetings,
Martin

[guyhaw]

My calculation (admittedly based on assumptions) and GAF both think there are only 80 wtm maxDTM positions.
There ought to be enough in the 96 positions we can see to tell us why FEG thinks there are more. We should be able to find one of them 'in another form' - and given that the wK is always on a1, the 'other form' has to be a flip around a1-h8. In fact, I haven't found the first or 96th positions duplicated, so maybe my 'divide by 24 for 4 Bs' idea is incorrect.
Yakov Konoval has DTC-based KBBBBK stats which I expect soon, but not I think DT-based ones.
g

[jshriver]

Hi all,

If you are interested in joining this effort, there are plenty of things
that remain to be done:

Good Morning,

I've computered some 5-1 egtbs using FEG several months back. If you want to email me a list of sets that need done I'm willing to both generate and share the data via my website. However I dont have net access at home right now. Hopefully by Jan/Feb things will settle down a bit for me. I miss getting to read this forums.

-Josh

[guyhaw]

I'd be interested in any information about the relationship between the .LOF and .LOG statistics, and the number of distinct (maxDTM) positions.
FEG seems to be throwing up, e.g., more maxDTM positions than I'd expected for KBBBBK. I did a simple 'divide by 24' and that was ok for btm maxDTM but appears not to be right for wtm maxDTM.
Also, examples of maxDTM positions, wtm and btm would be good.
Martin K can tell you which 5-1 FEG DTM EGTs he hasn't done: they'd be from the list KQBNNK, KQRNNK, KRBNNK and about 22 P-ful endgames.
g

[Martin Kreuzer]

Hi Josh,

nice to hear from you again. My computation of the 5-1 EGTB is in the final stages. I have already computed 6 more (kbbbpk, kbbnpk, kqbnnk, kqrnnk, krbbpk, krbnnk) and I am in the process of computing 2 more (kqrnpk and kbbppk). I will be posting notifications as soon as I get them uploaded. (Kirill's connection is not too fast.)

The remaining 17 EGTB will be computed over the Christmas vacation. Unfortunately, to help in this computation, you would have to download quite a few of them, because the more pawnful they get the more they interdepend.

I believe I will be finished in January. The next stage of the project would be to create meaningful statistics (Guy has already started this investigation) and check correctness via comparison to other computations (e.g. Guido's). The big next steps will be to figure out the position numbering scheme of FEG and to convert the files to other formats such as Nalimov's.

My best wishes for a joyful holiday season! I am looking forward to hearing from you again when your internet connection improves.
(A new batch of files on olympuschess would certainly be welcome by many EGTB friends!)

Greetings,
Martin

[jshriver]

My best wishes for a joyful holiday season! I am looking forward to hearing from you again when your internet connection improves.
(A new batch of files on olympuschess would certainly be welcome by many EGTB friends!)

Greetings,
Martin

Thanks, hope you have a happy holiday season as well.
I agree with the site, I feel sad to be neglecting it so, though I'm still surprised to see there is still a decent amount of downloads. I just started a new job about a month ago, moved in to a temporary place and my girlfriend/fiance and I are getting our first place together sometime around Jan/Feb. So lots going on.

Though I'm happy I finally found a good job (knocks on wood) so I can have more freedom in life.

Anyway I'm done being OT, hope everyone is well.
I'm going to try and give olympuschess a nice facelift in the next month or two, as well as a possible surprise

Take care,
Josh

[Martin Kreuzer]

Hi all,

now I have computed (and I am almost finished posting to egtb51) six new endgames:

kbbbpk, kbbnpk, kqbnnk, kqrnnk, krbbpk and krbnnk.

They are complete with stats files and md5sums. The total count of posted endgames is now 51 (out of 70). Three more endgames (kbbppk, kqrnpk and kqbnpk) have been computed already, but will be posted only in January when I am back from my Christmas vacation. By then I will have computed most of the remaining 16 5-1 EGTB.

I wish everybody a merry Christmas and a happy, healthy and successful New Year!

Greetings,
Martin Kreuzer

[jshriver]

Hi all,

I wish everybody a merry Christmas and a happy, healthy and successful New Year!

Greetings,
Martin Kreuzer

If you want I can lend you 55gig of online storage and make a ftp account to upload to olympuschess. Might make a dent in the 5-1 collection

-Josh

[guido]

... guido, for pointing out my finger-slip. Must have copied a cell down and forgotten to change the content - dangerous practice
All looks good for the correctness of your GAF EGTs. But do you have an independent way of verifying your EGTs?
Other maxDTM positions (at 'FEG' maxDTMs) will be interesting: no hurry.
g

Hi Guy, Martin and all,

my TBs and relative statistics have been always checked by me using gafs with option -v.
The cost in time is not trascurable but I do always this check after the generation to avoid that an error can propagate through the TBs.
The check is done comparing the congruency of the result of a given position with the results of the positions obtained executing all the legal moves starting from that position. This is done for each positions of both the files (wtb and btm). I think that this test is the same used by FEG, tbgen and also by MB.

Coming back to your excel table I saw that in the endgame KQQRRK (wtm) FEG indicates 17 positions in 4 moves while gafs only 10 positions in 4 moves. The difference in this case should be due to the symmetry in respect to the main diagonal. As a consequence I should found that of the 10 positions, 7 have this symmetry, but I found only 5 cases of symmetry.

The 10 positions are as follows:

8/8/8/4k3/8/8/RK6/QQR5 w - - - -
8/8/8/4k3/8/8/RK6/QRQ5 w - - - -
8/8/8/4k3/8/8/RK6/QR1Q4 w - - - -
8/8/8/4k3/8/8/RK6/QR5Q w - - - -
8/8/5k2/8/R7/8/RK6/QQ6 w - - - -
8/8/8/3k4/8/8/1RK5/QQR5 w - - - -
8/8/8/4k3/8/8/1RK5/QQR5 w - - - -
8/8/8/5k2/8/8/1RK5/QQR5 w - - - -
8/8/8/5k2/8/8/RRK5/Q1Q5 w - - - -
8/8/8/6k1/8/8/1RK5/QQR5 w - - - -

It is possible to know which are the 17 positions for FEG of tbgen?

Best wishes for a merry Christmas and an happy New Year to all.

Guido

[guido]

Hi Guy, Guido,

attached I post the max DTM positions for KBBBBK which
I extracted from the databases using FEG -LL

For wtm, the DTM seems to be 13 and there are more than 100 positions
(FEG lists only the first 100 or so)
For btm, the DTM seems to be 16 and there are 8 positions.

In the Christmas break, I will make more detailed experiments
(also with other endgames) and post the results here.

Greetings,
Martin Kreuzer

Thank you Martin.

I will examine your second attachment but it would be better if it was possible to have the list of the positions in FEN format, so we could easily compare FEG and gafs results.

My compliments for the generation of 5-1 TBs. I have problems in generating some endgames because are too big and use too much the disks. I should rewrite my generator .

Best wishes
Guido

[guyhaw]

Guido - I thought you were also eliminating unreachable positions in a more sophisticated way, where FEG does nothing like this.
So - there could be two unreachable positions as found by GAF.
I have ftp'd into the KK-hosted EGT51 website, so have got the latest stats. All 5-1 done, and 19 of 35 5-1p to go.
LOG/LOF continues to be exactly 2 for P-ful endgames, so I think the ratio is about reflections/rotations of the board. The ratios for P-less endgames continues to puzzle: maybe FEG is doing something clever when there are Bishops on the board?
g

[guido]

Guido - I thought you were also eliminating unreachable positions in a more sophisticated way, where FEG does nothing like this.
So - there could be two unreachable positions as found by GAF.
I have ftp'd into the KK-hosted EGT51 website, so have got the latest stats. All 5-1 done, and 19 of 35 5-1p to go.
LOG/LOF continues to be exactly 2 for P-ful endgames, so I think the ratio is about reflections/rotations of the board. The ratios for P-less endgames continues to puzzle: maybe FEG is doing something clever when there are Bishops on the board?
g

Hi Guy,

In the endgame KQQRRK GAFS cannot eliminate unreachable positions, because in the wtm there aren't checks. Moreover I (uselessly) repeated the calculation eliminating this control, obtaining the same result (10 positions in 4 moves).
The ratio 2 for P-ful endgames must be correct only if the number of pawns is odd. Differently there are positions which don't change with the symmetry and must be considered only once. The ratio for an even number of pawn should be < 2.
In similar but more complicated way this should happen also for P-less endgame where it is possible that some positions obtained by symmetry don't change in particular if there are like men.
This is only an hypothesis of mine. It is easy to be wrong.

Guido

[guyhaw]

Guido,

For P-less positions, the board can be rotated through 90 degrees three times to give 4 equivalent positions. The White King must be in some quarter of the board, so these 4 positions are all physically different.
It can also be reflected around some axis (e.g. a1-h8) to give (usually) four more positions. However, if the Kings are on a1-h8, the reflection might just be the same as before.
I think that LOF/LOF is the ratio between the number of physically-different positions and the number of essentially-different positions. Quite whether this is 8 (max), 4 (min) or somewhere in between rather depends on the actual set of maxDTM positions.
When there are Pawns on the board, the only transformation of the board allowable is a left-right 'flip', a <--> h. The flipped position is different as the wK is in the other half of the board. However, the position has not 'essentially' changed. Hence (I think) the LOG/LOF ratio of 2 (always) for P-ful positions.
I believe that neither the LOF or LOG figures take into account the fact that some men might be identical. So, if there are 'like men', both the LOG and LOF figures have to be divided by some factor - usually 2 but sometimes 4 (e.g. KQQRRK), 6 (KRNNNK) or 24 (KNNNNK). This can be checked by comparing the FEG stats for KPPK with the Nalimov ones.
The mystery of the KQQRRK maxDTM positions will be cleared up by the FEG data but I suspect there can be unreachable positions without the sntm King being in check.
g

[clocks]

Not sure what the problem is with my computer or the generator, etc.

I have FINALLY finished KQRRNK. I am hoping it is still of use, if so, let me know where to put it.

There is obviously something wrong with my system

Please do the rest yourself as this took over 2 weeks to complete. My apologies, I really need to setup a quicker system.

Derek

[guido]

Guido,

For P-less positions, the board can be rotated through 90 degrees three times to give 4 equivalent positions. The White King must be in some quarter of the board, so these 4 positions are all physically different.
It can also be reflected around some axis (e.g. a1-h8) to give (usually) four more positions. However, if the Kings are on a1-h8, the reflection might just be the same as before.
I think that LOF/LOF is the ratio between the number of physically-different positions and the number of essentially-different positions. Quite whether this is 8 (max), 4 (min) or somewhere in between rather depends on the actual set of maxDTM positions.
When there are Pawns on the board, the only transformation of the board allowable is a left-right 'flip', a <--> h. The flipped position is different as the wK is in the other half of the board. However, the position has not 'essentially' changed. Hence (I think) the LOG/LOF ratio of 2 (always) for P-ful positions.
I believe that neither the LOF or LOG figures take into account the fact that some men might be identical. So, if there are 'like men', both the LOG and LOF figures have to be divided by some factor - usually 2 but sometimes 4 (e.g. KQQRRK), 6 (KRNNNK) or 24 (KNNNNK). This can be checked by comparing the FEG stats for KPPK with the Nalimov ones.
The mystery of the KQQRRK maxDTM positions will be cleared up by the FEG data but I suspect there can be unreachable positions without the sntm King being in check.
g

Hi Guy,

Sorry for my error. The ratio LOG/LOF for P-ful endgames must be 2 in any case independently by the parity of the number of the pawns because in any case the wK and the bK are single.

For P-less endgames the things are different.
Tbgen and gafs keep into account the 8 symmetry by putting one K in the triangle a1-d1-d4. Gafs puts always (in wtm and btm cases) the wK in the triangle while it seems to me that tbgen does the same with the King to move.
You surely know better than me if this last info is true.
Apart this detail, with this choice all the symmetry are considered except the case when the two kings are on the main diagonal; for this case gafs, differently from tbgen, set as illegal one of the two equivalent positions.

On the contrary FEG starts considering all the positions and successively keeps into account the symmetry. But not always there are 8 equivalent positions for any independent configuration because of the two symmetry in respect to the main (a1 - h8 ) and the_secondary_diagonal (h1 - a8 ). Such cases happen when the kings are on one of these diagonals while the other pieces are:

- On the same diagonal of the kings
- In symmetrical positions two by two
- In both the preceding situations

So again, if I'm not wrong, the ratio should be always less than 8.
The question could be: does FEG keep into account the symmetry in respect to the diagonals? If the results of FEG and tbgen coincide, should FEG keep into account only the symmetry in respect to the secondary diagonal and not to the main diagonal?

A good question for Christmas holidays

Ciao a tout le monde
Guido

P.S. I read now your answer. Yes, there is also surely to consider the problem of like pieces as you say, but my point "- On the same diagonal of the kings" holds in any case independently of like pieces, at least it seems to me.
Therefore should the ratio be always less than 8*n! if there are n like pieces and analogously in other cases?

Greetings again

[guyhaw]

Guido,

You are correct in saying that the parity of the number of Pawns does not affect FEG's LOG/LOF ratio. Because there is only one side-to-move King, LOG/LOF is determined largely by how many place the stmK can be rotated/reflected to. With Pawns on the board, rotations are not allowed, so LOG/LOF is exactly 2, or should be.

TBGEN knows about 'stm' and 'sntm' rather than 'White' and 'Black'. For P-less positions, it has the stmK in a1-d1-d4, and if the stmK is on a1-d4, the sntmK is in a1-h1-h8. This still leaves the case of both Kings on a1-h8, but TBGEN makes no further restrictions (as Christoph Wirth's code and GAF do). I think the simplest way to exclude one of two equivalent positions is to set the position with the higher index to 'broken'.

The FEG LOG/LOF ratio for the count of maxDTM positions does in fact vary (for P-less endgames) from exactly 4 (e.g. KBBNNK) to exactly 8 (e.g. KBBBBK). Oddly, for KBBNK it is exactly 8 with wtm and 4 with btm: I guess there is something I don't know about FEG's indexing system.
So, the already-published statistics show that it is not always less than 8.

FEG will treat the symmetry about the 'main diagonal' as you call it (a1-h8?) and the 'secondary diagonal' [ditto] (a8-h1?) in the same way, because a 90-degree rotation moves 2 Ks on a1-h8 to 2 Ks on a8-h1. How FEG treats the symmetry can be checked by looking at the FEG stats for small P-less endgames, e.g. KQK, KRK, KQKR etc.
g

[guido]

Guido,

You are correct in saying that the parity of the number of Pawns does not affect FEG's LOG/LOF ratio. Because there is only one side-to-move King, LOG/LOF is determined largely by how many place the stmK can be rotated/reflected to. With Pawns on the board, rotations are not allowed, so LOG/LOF is exactly 2, or should be.

TBGEN knows about 'stm' and 'sntm' rather than 'White' and 'Black'. For P-less positions, it has the stmK in a1-d1-d4, and if the stmK is on a1-d4, the sntmK is in a1-h1-h8. This still leaves the case of both Kings on a1-h8, but TBGEN makes no further restrictions (as Christoph Wirth's code and GAF do). I think the simplest way to exclude one of two equivalent positions is to set the position with the higher index to 'broken'.

The FEG LOG/LOF ratio for the count of maxDTM positions does in fact vary (for P-less endgames) from exactly 4 (e.g. KBBNNK) to exactly 8 (e.g. KBBBBK). Oddly, for KBBNK it is exactly 8 with wtm and 4 with btm: I guess there is something I don't know about FEG's indexing system.
So, the already-published statistics show that it is not always less than 8.

FEG will treat the symmetry about the 'main diagonal' as you call it (a1-h8?) and the 'secondary diagonal' [ditto] (a8-h1?) in the same way, because a 90-degree rotation moves 2 Ks on a1-h8 to 2 Ks on a8-h1. How FEG treats the symmetry can be checked by looking at the FEG stats for small P-less endgames, e.g. KQK, KRK, KQKR etc.
g

Hi Guy,

About P-ful endgames we agree completely. There is only one vertical symmetry and the men cannot stay on the axis of symmetry, as happens for diagonal symmetry. So the value of 2 for the ratio (like men apart) would seem compulsory.

About the positions of the two kings for p-less endgames tbgen and gafs coincide because both use 462 combined positions. What could be different is the order of these positions and the algorithm to access the tables (gafs uses tables). Still gafs sets to broken the position with higher index if the kings are on the diagonal a1-h8.

What you say about the value of the ratio between 4 and 8 for P-less endgames is easily explainable even if a check should be done in particular cases.
When I said that the ratio is always less than 8, in my mind I thought to the total number of positions of the endgame and not only to the maxDTM!

In fact let us consider these two positions in the KRK endgame:

1. wKa1, wRa2, bKc3
2. wKa1, wRb2, bKc3

The position 1 represents one of the 8 equivalent positions, while position 2 one of the only 4 equivalent positions.
Therefore as in KRK endgame there are both these types of positions the weighted mean will be always inside the interval 4-8.

In endgames with like pieces the positions symmetrical to the diagonal axes are still of type 2 (ratio 4)

But your calculations of the ratios is referred only to maxDTM and it is possible that in the cases with ratio 4 or 8 all these positions are only of one of the two mentioned type. In fact when the numbers are high it is more likely that the ratio is different from the extreme values, as appears from your table.

In KBBBBK with ratio 8 there are 80 positions for wtm and it seems that all these positions don't have both the kings on a diagonal. The Martin's list contains more than 80 positions, all with the wK in a1 and some with the bK in the upper triangle. This would explain the difference with your value of 80 where you speak of distinct positions.
Probably the Martin's list contains duplicates by symmetry in respect to the diagonal a1-h8.

It would be interesting to see if in the cases of ratio 4 all the positions are of type 2.

Ciao
Guido

P.S. If I write in bad English, pls correct me freely.

[guyhaw]

Guido - your reasoning is correct and clear: I think we have both been distracted at times by thinking about all positions rather than the set of maxDTM positions.
There are 1806 distinct positions of the Kings with Pawns, and 462 with no Pawns. The ratio is 3.91 rather than 4 because of the 'two Kings on long diagonal' subtlety. So I would expect to see LOG/LOF ~ 7.82 for 'all positions' and maybe the FEG statistics reveal this.
As your KRK example neatly shows, some positions are in a 'set of 8' but diagonally-symmetric ones are in a set of 4.
As you say, Martin includes maxDTM positions with the bK above a1-h8, and this was avoidable - though not perhaps with FEG. One suspects this is the reason for duplication as you say, but I have not found two positions mirroring each other in a1-h8. However, I have not tried very hard and probably will not.
An easier way to see if FEG is reporting maxDTM positions in diagonally-mirrored pairs is to have it provide the maxDTM positions for a smaller P-less endgame with fewer maxDTM positions: KQK with 1 wtm maxDTM positions suggests itself. KBKN has 2 wtm and 1 btm, KQKQ has 2 and 2, KRKN 1 and 1 and so on - see my stats spreadhsheet at http://www.icga.org - Western Chess, Endgame data.
g

[guido]

Guido - your reasoning is correct and clear: I think we have both been distracted at times by thinking about all positions rather than the set of maxDTM positions.
There are 1806 distinct positions of the Kings with Pawns, and 462 with no Pawns. The ratio is 3.91 rather than 4 because of the 'two Kings on long diagonal' subtlety. So I would expect to see LOG/LOF ~ 7.82 for 'all positions' and maybe the FEG statistics reveal this.
As your KRK example neatly shows, some positions are in a 'set of 8' but diagonally-symmetric ones are in a set of 4.
As you say, Martin includes maxDTM positions with the bK above a1-h8, and this was avoidable - though not perhaps with FEG. One suspects this is the reason for duplication as you say, but I have not found two positions mirroring each other in a1-h8. However, I have not tried very hard and probably will not.
An easier way to see if FEG is reporting maxDTM positions in diagonally-mirrored pairs is to have it provide the maxDTM positions for a smaller P-less endgame with fewer maxDTM positions: KQK with 1 wtm maxDTM positions suggests itself. KBKN has 2 wtm and 1 btm, KQKQ has 2 and 2, KRKN 1 and 1 and so on - see my stats spreadhsheet at http://www.icga.org - Western Chess, Endgame data.
g

Hi Guy,

First of all my compliments for the very interesting and useful endgame data table, and also for your papers on chess, which I hope to read in future.

Intuitively your evaluation of 7.82 for the mean value would seem correct.

But it is possible to do the exact computation in the very simple KRK endgame.
The number of total positions (legal + illegal) of the endgame is:

64*63*62 = 249,984

The number of total positions with ratio 4 are those in which the 3 men are all on each of the two diagonals:

(8*7*6) * 2 = 672

Therefore 249,984 - 672 = 249,312 should be the total positions with ratio 8.

The total distinct positions are:

249,312/8 + 672/4 = 31,164 + 168 = 31,332

Finally the weighted ratio will be:

249,984 / 31,332 = ~ 7.97855

The result doesn't match with 7.82 but it is possible that my computation is wrong.

About the problem of the positions in wtm KBBBBK endgame I examined the first position of the Martin's list where bK is in the upper triangle.

Ka1, Bc1, Bc6, Bd2, Be1, kd6

My program gives a win in 11 moves (not 13!) with Bc6-b4 or Bc6-d4.
Can you check this result?

Ciao
Guido

[guyhaw]

You're welcome re the ICGA .xls and the papers: re the papers, just skip the maths if you wish and go to the end results.

For your KRK calculation, I would do it a different way
- the total number of positions = 3,612 (for Kk positions) * 62 = 223,944
- the total number symmetric on a1-h8 (and a8-h1) = 42*6*2 = 504
- the number of non-a1-h8-symmetric is 223,440, in sets of 8 being 27,930 distinct positions
- the number of symmetric distinct positions = 504/4 = 126
- the number of distinct positions = 28,056
- 223,944/28,056 = 7.9820, still greater than my 7.82
g

[guido]

<cut>

About the problem of the positions in wtm KBBBBK endgame I examined the first position of the Martin's list where bK is in the upper triangle.

Ka1, Bc1, Bc6, Bd2, Be1, kd6

My program gives a win in 11 moves (not 13!) with Bc6-b4 or Bc6-d4.
Can you check this result?

Ciao
Guido

Hi Guy,

sorry, I confuse bishops with knights.
The position is won correctly in 13 moves. Therefore the Martin's list contains duplicates.

Guido

[guido]

You're welcome re the ICGA .xls and the papers: re the papers, just skip the maths if you wish and go to the end results.

For your KRK calculation, I would do it a different way
- the total number of positions = 3,612 (for Kk positions) * 62 = 223,944
- the total number symmetric on a1-h8 (and a8-h1) = 42*6*2 = 504
- the number of non-a1-h8-symmetric is 223,440, in sets of 8 being 27,930 distinct positions
- the number of symmetric distinct positions = 504/4 = 126
- the number of distinct positions = 28,056
- 223,944/28,056 = 7.9820, still greater than my 7.82
g

Hi Guy,

I agree with your evaluations, because are coherent with the ratio 7.82 obtained keeping into account only the legal positions of the two kings, while my computation considered all the positions, legal and illegal, of the kings.

Nevertheless apart the cases where the ratio is 8, it is strange that all the values you found for maxDTM are far from the mean value (7.82 or 7.98 ) and shifted to 4, as there was an high percentage of positions with ratio 4
much higher than the global mean of the endgame. This could mean that in general in 5-1 TBs diagonally symmetrical positions are farther from the mate than the others.

Guido

[Martin Kreuzer]

Hi Derek,

congratulations on generating kqrrnk. Indeed, as you know form the messages I posted further up, I have computed this EGTB quite some time ago. It would be interesting, however, if you could compute the md5sums of your files and check them against the md5sum I posted on egtb51. In this way we'd have some check on the correctness of the files I posted.

Unfortunately, the checks for correctness built into FEG are very superficial: the program verifies only that the files decompress correctly. A more thorough way of verifying the results of the computation would be useful.

Cordial greetings and best wishes for the holiday season,
Martin