H.G.Muller

Welcome!

Hello, H.G.Muller, and welcome to Wikipedia! Thank you for your contributions. I hope you like the place and decide to stay. Here are a few links to pages you might find helpful:

• Introduction and Getting started
• Contributing to Wikipedia
• The five pillars of Wikipedia
• How to edit a page and How to develop articles
• How to create your first article
• Simplified Manual of Style

Please remember to sign your messages on talk pages by typing four tildes (~~~~); this will automatically insert your username and the date. If you need help, check out Wikipedia:Questions, ask me on my talk page, or click here to ask for help here on your talk page and a volunteer will visit you here shortly. Again, welcome!

Your tenjiku shogi page

Hello, I saw your tenjiku shogi page on your website. Maybe the fire demon should slide to the sides, instead of forward and backward? Perhaps both rule sets could be supported.

The tenjiku shogi article says that the Edo-era sources disagree on this, and that Western sources have largely used the file interpretation: but using the rank interpretation means that it keeps the movement of the water buffalo, in keeping with it being a promoted water buffalo.

(There are other disagreements for tenjiku, but I gather most of them are the Western sources going against the Edo-era ones, and the Edo-era ones seem definitely more trustworthy.) Double sharp (talk) 16:48, 26 June 2014 (UTC)

I wrote a long post on the issues in creating a ruleset for tenjiku shogi based on the original sources at Talk:Tenjiku shogi#Controversies over piece moves. But usually it seems to be as I said above: some parts of the Japanese rules might have gotten lost in translation when the TSA started making sets for the LSVs, perhaps? Double sharp (talk) 16:58, 26 June 2014 (UTC)

Some parts of that post did kind of get distracted into working out the values of the tenjiku shogi pieces, though. Double sharp (talk) 17:02, 26 June 2014 (UTC)

OK, I just discovered that page, and posted a remark on the jump-capturing generals before I saw your remark here. Recently I met someone who had known George Hodges personally, and he told me that in compiling the rule leaflets, George was in the habit of making arbitrary decisions to resolve ambiguities or inconsistencies between sources concerning the rules of these (often scarcely documented) games. He considered it of primary importance that everyone would accept the same rules, and shielded the public from any controvercy. The article on my website was mainly based on the Wikipedia description (which did not mention the controversy about the Fire-Demon move at that time), plus the conclusions I had been drawing myself from the discussion conducted by Colin Adams about the viability of the rules as spread by TSA. I will probably change it to conform to progressing insight. The main reason for posting my Tenjiku article was as supporting info for my program 'HaChu' for playing large Shogi variants. I am still working on implementing Tenjiku Shogi there, and until I do I can still change the rules as seems best. H.G.Muller (talk) 21:23, 26 June 2014 (UTC)

OK, thank you. Thanks to your info, my opinion of the TSA's actions with regard to the LSVs has plummeted even further. I always thought they should have specifically annotated all disputed moves by showing the readings of the Japanese sources, and let the readers make up their own minds, kind of like what you see in urtext editions for classical music. Of course they could still treat one as primary. But instead they suppressed the variations and sometimes ended up making up even more variant moves, and then proceed to claim that proposed changes were invented and not historically supported even when they actually in all likelihood reverted back to the original Japanese version (lion hawk from tenjiku shogi). Double sharp (talk) 06:20, 27 June 2014 (UTC)

I also replied to your comment at Talk:Tenjiku shogi: but I still have a few questions regarding the rules. One of them is on that page. The other one is: how would you interpret the situation where a fire demon lands next to an enemy fire demon? TSA states that only the moving piece is burnt: some other computer programs and books say that all adjacent pieces are also burnt, apart from the enemy fire demon: Japanese Wikipedia confirms that the moving piece is burnt, but says nothing regarding the fate of other adjacent pieces. Double sharp (talk) 06:42, 27 June 2014 (UTC)

The TSA rule that the moving Fire Demon dies without affecting its surroundings seems quite natural to me. If the moving Demon does not manage to burn the static Demon, why should it burn any other pieces? The fact that Demons get fried just like any other piece when they approach an opponent Demon shows they are not particularly resistant to burning, but burn just as easy as other pieces. H.G.Muller (talk) 16:25, 27 June 2014 (UTC)

I agree with you. So I assume that this must mean that the passive burn must happen immediately after the opponent's move finishes, but before any effects of that move (IIRC the only effect is the fire demon's burn) occur.

What happens if a lion mover or area mover passes (but does not stop) next to a fire demon? Will it be immolated? (I think not.) Double sharp (talk) 11:47, 1 July 2014 (UTC)

For questions like this we have little else to go by than Occam's razor, as the historic rule descriptions do not specify such minor rule details. My reasoning would be that such transient encounters do not count, because when the Fire Demon is the moving piece that makes an area move, it also only burns the neighbors of its final square, and not those of the squares it passes through. Similar for the sliding moves of the Demon, and moves of other sliders that pass through the burn zone. Area moves are really not much different from slider moves; they are not restricted to always step in the same direction, but that is all. H.G.Muller (talk) 12:11, 1 July 2014 (UTC)

Talkback

Hello, H.G.Muller. You have new messages at Double sharp's talk page.
You can remove this notice at any time by removing the {{Talkback}} or {{Tb}} template.

Double sharp (talk) 11:35, 26 September 2014 (UTC)

Replied again. Double sharp (talk) 03:32, 29 September 2014 (UTC)

Promotion rules for dai and tenjiku shogi

Are they the same as in chu shogi (Chushogi Renmei rules)? Double sharp (talk) 05:52, 2 December 2014 (UTC)

In tenjiku this is what I would expect:

• Promotion is permanent.
• The king, great general, vice general, free eagle, lion hawk, and fire demon do not promote.
• Pieces may not promote more than once.
• When a piece enters the zone through moving or capturing, it may choose to promote or defer promotion.
  • If it defers, it may not promote unless:
    • it makes a capturing move that either starts in the zone, ends in the zone, or both; or
    • after leaving the zone, it reenters either through moving or capturing, when it may choose to promote or defer promotion again.
• A pawn, iron general, or lance reaching the final rank becomes immobile. A knight reaching the second-last rank also becomes immobile.

We might thus get closer to having a standard rule set for the other less-played large variants (of which tenjiku is surely among the best IMHO).

(I'll probably post more at Talk:Tenjiku shogi.)

Double sharp (talk) 06:14, 2 December 2014 (UTC)

This sounds very much like the rules most likely would have been. But I don't think we will ever know for sure; the information has been lost over time, if it was ever written up in such detail. Tenjiku seems clearly derived from Chu, and thus would have inherited most of its rules. That it did not inherit the Lion-capture rules is logical, as the Lion is just 'one of the crowd', not having the status of the all-powerful game-chnging piece it is in Chu. In Tenjiku that is the Fire Demon, you start with two of them, and the game would be quickly finished if they would not quickly disappear from it after doing tremendous damage. Note that the area movers cannot promote anyway, so the only issue with passing through the zone is with Lion, Eagle and Hawk. I do agree that it is slightly more logical to not promote on an in-out move. A more tricky point is whether there should be an exception for irreversible pieces. There never is a reason to defer promotion for Lance or Pawn, but Knight and Iron promote to pieces that are not upward compatible. But the question can still be asked: if for tactical reasons you decide not to promote N or I, should they really get a second chance, or is it reasonable to require you just 'write them off'? You still have a chance to promote them through capture, and they are more mobile than a Pawn, and the zone is deeper. So I can live with the rules you give here. And it seems a moot point anyway, as N and I in Tenjiku are like sand grains in an asteroid storm, and have no hope to ever reach last rank. H.G.Muller (talk) 11:31, 30 January 2015 (UTC)

You're right, I forgot that the promotions of N and I are not upward-compatible. Since there can be a reason to defer promotion, I think I should alter the rule to be the same as that of pawns in chu shogi: if you decline the promotion, you get another chance (which you cannot refuse) when you reach the end of the board (for the knight, the second-last rank). But really it is a moot point, as you eloquently put it. Double sharp (talk) 14:20, 30 January 2015 (UTC)

mating potentials on large boards

You mention here that King+Crown Prince can force checkmate on 14×14, but not on larger boards (so not Tenjiku), and that King+Gold can force checkmate on 10×10 but not larger. Why is this so? It's a very interesting result. Double sharp (talk) 06:45, 2 December 2014 (UTC)

It is quite common that weak short-range leapers lose their mating potential beyond some board size. When they are not powerful enough to force the bare King towards an edge (like a Lion can do by checking, and a KN (Centaur) can do by zugzwang), a succesful defense is just to run away from the attacking King. The latter can only just keep up with you, but then there would be no opportunity for the other piece to move at all. For each move the other piece does, the attacking King loses again one square on you. Some pieces (like Gold) can obstruct the bare King by opposition, so that at least it does not increase the distance to the other King, but to position it such takes at least one move, so you still cannot gain on it with that other King. Whether mates can be forced on arbitrarily large boards depends on the minimal group of pieces that you need to meaningfully obstruct the bare Kings escape to freedom, and the speed of the group as a whole (e.g. a pair of Knights only can advance 1 square per turn if they move alternately, and thus cannot catch up with the bare King. So King plus an arbitrarily large number of Knights will always get in trouble if the board is sufficiently large. The smallest group that moves faster than King is just a single Knight, and a single Knight cannot slow down a King at all.) The actual maximum board size for K + +E and K + G was obtained by generating tablebases for those end-games, and looking at the percentage of won positions in them. H.G.Muller (talk) 11:53, 30 January 2015 (UTC)

micro-Max improvement suggestion (12 characters)

Hi Harm,

I'd like to suggest a modification to micro-Max that can cut 12 characters without affecting play. I'm considering this version.

The penalty for off-center squares is given as

(i-4)*(i-4)+(j-3.5)*(j-3.5)

This is equivalent to

$(i-4)^{2}+(j-3.5)^{2}=i^{2}-8i+16+j^{2}-7j+12.25=i^{2}-8i+j^{2}-7j+28.25$

The scores are only used by taking the difference of scores between two squares, so the constant term is cancelled out and is irrelevant. This means that the scores are equivalent to $i^{2}-8i+j^{2}-7j$, which can be written as

i*i-8*i+j*j-7*j

Which is 16 characters, instead of 28.

On the same note, I don't understand why you have $(i-4)^{2}$ instead of $(i-3.5)^{2}$. This way, you're not penalizing for being away from the center, but rather for being away from the 5th rank. This means 4th and 6th rank have the same penalty despite 4th being closer to the center, and the penalty for the 1st rank is huge. If you change it to $(i-3.5)^{2}$, the above becomes

i*i-7*i+j*j-7*j

Which can also be written in several other ways, of equivalent length:

i*(i-7)+j*(j-7)
i*i+j*j-7*(i+j)

-- Meni Rosenfeld (talk) 18:04, 28 February 2015 (UTC)

Draft:Betza's funny notation concern

Hi there, I'm HasteurBot. I just wanted to let you know that Draft:Betza's funny notation, a page you created, has not been edited in 5 months. The Articles for Creation space is not an indefinite storage location for content that is not appropriate for articlespace.

If your submission is not edited soon, it could be nominated for deletion. If you would like to attempt to save it, you will need to improve it.

You may request Userfication of the content if it meets requirements.

If the deletion has already occured, instructions on how you may be able to retrieve it are available at WP:REFUND/G13.

Thank you for your attention. HasteurBot (talk) 01:32, 8 January 2016 (UTC)

Your mnemonic piece sets

Can I upload pictures of your mnemonic piece sets being used for the large shogi variants (chu, dai, tenjiku, dai-dai, maka-dai-dai, tai) that come with WinBoard Alien Edition? (All initial positions.) Double sharp (talk) 12:15, 8 February 2016 (UTC)

Sure, no problem. I don't know if you already had a place in mind where they are on-line (I replaced them by 'Interactive Diagrams' in the chessvariants.org articles, although the original PNG images are still there for display in a JavaScript-disabled browser), but if not you could just make screenshots of WinBoard (don't forget to tick the 'flip black' option, though, or some of the 'custom designs' like Knight would be displayed upside-down for black). Note that in these new images I started using a short dash close to the center of the piece as encoding for a range-5 move in these new images, where the WinBoard Alien Edition still uses a dot closer to the center than the dot used to indicate range-3. (Which struck me as illogical for a longer-range move.)

Ah! You're right. (I originally did the WinBoard screenshots, but forgot to tick "flip black": I'll fix it. They're not uploaded yet.) But those new images won't give tai shogi, while WinBoard screenshots would. Double sharp (talk) 14:06, 8 February 2016 (UTC)

Betza's Atomic Theory

I reread Betza's atomic theory, and listed the possible pieces without introducing modifiers. Since he states that 2 atoms = minor piece, 3 = rook, 4 = princess, 5 = queen, 6 = rook + nightrider, 7 = amazon, I assume he thinks the nightrider is also worth 3 atoms.

1. Atomic leapers:
   • W, F, D, A (atomic leapers)
2. Minor pieces:
   • WF, WD, WA, FD, FA, DA (fairy minor pieces; two atoms)
   • N, B (orthodox)
3. Rooks:
   • WFA, WDA, WFD, FAD (three atoms)
   • NW, NF, ND, NA (knight + atom)
   • BW, BD (bishop + atom)
   • R (orthodox)
   • NN (nightrider)
4. Princesses (notwithstanding that the princess herself needs to be moved up a category):
   • WFAD (four atoms)
   • NWF, NWD, NWA, NFD, NFA, NDA (knight + two atoms)
   • NB (orthodox)
   • RF, RA (rook + atom)
   • NNW, NNF, NNA, NND (nightrider + atom)
5. Queens
   • NWFA, NWDA, NWFD, NFAD (knight + three atoms)
   • NBW, NBD (princess + atom)
   • RFA (rook + two atoms)
   • Q, RN (orthodox)
   • NNWF, NNWD, NNWA, NNFD, NNFA, NNDA (nightrider + two atoms)
   • NNB (princessrider)
6. Empressriders
   • WFADN (knight + four atoms)
   • NBWD (princess + two atoms)
   • RNF, RNA (empress + atom)
   • NNWFA, NNWDA, NNWFD, NNFAD (nightrider + three atoms)
   • NNBW, NNBD (princessrider + atom)
   • NNR (empressrider)
7. Amazons
   • QN (orthodox)
   • RNFA (empress + two atoms)
   • NNWFDA (nightrider + four atoms)
   • NNBWD (princessrider + two atoms)
   • NNRF, NNRA (empressrider + atom)
8. Amazonriders
   • QNN (amazonrider)
   • NNRFA (empressrider + two atoms)

(I label the Capablanca-like compounds as orthodox as they are composed wholly of orthodox pieces.)

Just how close is this theory to being right? If it's mostly right, with a few outliers like the princess which needs to be moved up a category, then it's a good source of rough estimates that need to be refined further through accounting for other effects. If it's mostly wrong, something is wrong with the whole theory. (This whole ranking thing is somewhat reminiscent of pocket mutation chess, so it might make a good experimental basis for that game.)

I felt free to exclude cases like RD, a clearly weak enhancement to the rook, because Betza noted that this sort of case would of course not count as much as RF or RA would. (Is this true?)

Not all of the pieces in each class need to be tested, of course (that's just a wonderful dream), but imagine many would have already been. Double sharp (talk) 14:33, 8 February 2016 (UTC)

This is reasonably close to being right. Long time ago I tested many of the short-range leapers (i.e. combinations of W, F, N, A, and D). The conclusion was that the values correlated quite well with the formula 1.05*(30 + 5/8*N)*N (centi-Pawn), when N was the number of move targets. Values of pieces with the same N can differ because of 'emergent' properties like 'speed', 'color binding', 'mating potential' etc. So an Alfil will definitely be worth much less than a Ferz, despite the fact that they both have 4 moves, because of the much higher degree color binding. But in general this only gives a minor modification on the value predicted by the formula.

Sliders are more tricky. I still did not find a satisfactory explanation why the BN is so unexpectedly strong. But given that it is, I would expect a similar upgrade for NNFA and NNB. As to the value jumping, the FA appears to be worth about the same as B, so apparently the power to jump rather than slide to the second square is worth as much as all more-distant moves together. But that did not hold for Queen; there the ability to jump to 2nd was only worth as much as the moves further than 3 steps. I never tested this on the Rook. I never tested systematically how large a fraction of the normal value of the atom remains when the move is blockable, but I suspect that this would also depend on whether it is blockable on a square that the piece itself attacks, or not. The Xiangqi horse is a very bad example of the latter, as you can block two of its moves on a single unattacked square. It is only worth half a Knight.

With modifiers, forward moves appear to be about worth twice as much as backward or sideway moves, and captures about twice as much as non-captures.

P.S. I am looking for some feedbeack on my draft article on Betza notation, mentioned above. Could you hve a peek at it? H.G.Muller (talk) 16:27, 8 February 2016 (UTC)

I had a look at it. It looks very good (and given that your extension is published, there should not be any objections to it), especially for such a difficult-to-explain topic in some cases. There are certainly enough examples for me to get the idea. In fact, perhaps we should roll it out over all the large shogi variants (except taikyoku which I still adamantly refuse to touch).

I'm guessing that one reason why the princess is so strong is that her power does not tail off as sharply as the RA or RF. If we look at the RF, it has the concentrated power of a prince (commoner) right next to it, but after that it has the power of a rook: able to interdict the enemy king, but no more than that. However, if we look at the princess, right next to it it acts like a ferz, but its concentration of power actually increases in the middle of her range because the knight moves end up right next to the bishop moves, creating 2×2 solid blocks of attacked squares. After that she acts like a bishop. Thus she can attack pawn chains or pieces from a distance (so that she is harder to chase away) while still maintaining a superpiece-like concentration of power (like the queen or empress), whereas the dragon king really has to be next to its prospective victims to have an effect.

A princess or an empress on a 16×16 board (like Chess on a Really Big Board) would be interesting. From a distance, the princess would attack like a bishop from far away, slicing through the pawn chains. But now she really does need to get relatively closer to attack, and given the presence of spaciously large leapers and leaper-riders like the superknight (NLJ) and rose, it becomes easier to drive her away. She is now obviously far weaker than the queen, but still a force to be reckoned with. Another effect is that on an 8×8 board, the different directions of the queen move are still rather close. On the 16×16 board, they can diverge very far apart, and while this may be useful in creating forks, it also becomes more difficult to attack one thing from multiple directions.

P.S. Given my fondness for large variants, I wonder if it is possible to find out the approximate values of pieces in Chess on a Really Big Board? (On a 16×16 board, we have the six orthodox chess pieces, and also the FD, WFA, NLJ, princess, empress, and rose.) I suspect the rose may cause some trouble, though. Double sharp (talk) 04:11, 9 February 2016 (UTC)

There do not exist many programs that can handle such a large board; most configurable programs are limited to boards of 128 squares. The past month I upgraded Fairy-Max to get rid of the 8-rank limitation, and internally it now has a 16x16 board. But it needs a 'guard band' the sizes of the largest sideway leap to prevent pieces jumping 'across the board edge' to the next rank, effectively kaming it a (slanted) cylinder board. So with nothing worse than Knights the maximum width is 14. (But 14x14 would of course already be interesting too.) There are many aspects of CoaRBB that it currently cannot handle, though. Castlings cannot move the King more than 4 squares, and initial Pawn pushes cannot go further than 5 steps, and it could then only be captured on the last square it skipped. HaChu can handle much bigger boards, but being a Shogi program castling and e.p. capture are also problematic there, and the evaluation is not really suitable for Chess. (In many aspects HaChu is still in its infancy.) Of course for determining piece values such details might not really matter, or could be circumvented by using a judiciously chosen initial position. (E.g. where the Kings already start near a corner, and many of the Pawns in more advanced locations.) Board filling fraction is also a concern; when the potential range of sliders is so large it could have much more impact than on 8x8. So it would be a pretty large project, and it is not likely I will be able to tackle it any time soon. Much closer to the top of my to-do list is making HaChu able to play Macadamia and Cashew Shogi. This will pose a similar problem, however, as virtually nothing is known about piece values there either. I am not even sure I can make HaChu play a realistic strategy for Macadamia, w.r.t. which pieces should lead the attack, and which pieces should support it and which should remain behind for defensive purposes. The key in that game seems to be getting favorable promotions. In Chu Shogi finding a good strategy w.r.t. advancing steppers vs developing sliders proves already a big problem. Without a realistic strategy empirically measured piece values cannot be trusted.H.G.Muller (talk) 08:18, 9 February 2016 (UTC)

I see. Well, if it's a large project that's far down on your to-do list, I won't mind waiting some time. Nonetheless, perhaps it would be possible to figure out piece values on 16×16 indirectly, by looking at 8×8, 10×10, 12×12, and 14×14, and then trying to see if some sort of trend is at work. To that end it seems the CoaRBB pieces form a pretty good set: we have infinite-range riders (B, R, Q); finite-range riders (Rose); mixed (BN, RN); and leapers (NLJ, N, FD, WFA), as well as the ubiquitous K and P. (Maybe a workaround would be to accept later contact between the sides, and use only two-step pawn moves; but I wonder how much of the pawn's opening value comes from the fact that the central pawns' first opening moves already stake claim to central territory. 1.i4 in CoaRBB certainly helps development, but doesn't claim the centre yet, while 1.i8 does – with the problem that you do not yet control the intermediate territory from the fourth to eighth ranks. If we used short pawn moves, then each side expands slowly, and when they meet, they have fully claimed the space behind them. I don't know how much of a difference this would make.)

I am not very clear on proper chu shogi strategy either. Going by "chessplayer's intuition", I would expect that an unsupported attack by one's strongest piece (the lion) should be beaten back with loss of time, perhaps by advancing pawns to let the dragon horses out to harass the opponent's lion so that it wanders around without doing anything. But that is not what I see people doing in online games. Instead the lions get developed first, seek each other out, and form a moat of death between them. (Well, death for steppers; riders can pass through.) I wonder how much of this is because of the prohibition on trading lions (which I always felt to be a rule for the sake of a rule; since tenjiku functions perfectly well even after the fire demons are burned off, and chess perfectly well even after the queens are exchanged, I don't see the reason for it). If lions could be traded, an immediate answer would be to trade the lions once they form this bridge of death. As it stands, though, since the lions are patrolling in the centre, one has to keep watch there, and any direct contact between lesser pieces gets carried out through edge attacks trying to force an incursion to the promotion zone. (To whoever said tai shogi is like a series of local skirmishes: I agree, but it seems that chu is like that already!)

Still, I would expect that basic chess principles should still hold, except for things related to the FIDE pawn. (Which would actually not make a bad inclusion in chu shogi, as FIDE pawns can defend better against a lion than shogi pawns.) Double sharp (talk) 14:17, 9 February 2016 (UTC)

I think the anti-trading rules for Lions serve a really important function in Chu. Even Chess loses most of its glamour after trading Queens, but in Chu it would be much worse due to the larger number of remaining pieces. And it is much easier to avoid Queen trading in Chess than it would be to avoid Lion trading in Chu. I haven't watched any Tenjiku games, but I imagine there is much less need to protect the pieces there. Fire Demons are very different in character from Lions. The latter derive their value from doing a lot of accumulated damage over the duration of the game. The Demons would typically do all their damage in a single kamikaze action (which still amounts to more than what a Lion could do in the entire game. So you can trade a Demon for 3 or 4 Queen-class pieces, and much later in an unrelated event the opponent would trade his Demon for 3 or 4 of your pieces. This is much easier than trying to trade the Demons for each other, and it is also not clear how it could ever be prevented by any rules. With the Lion getting an equal trade for non-Lion material is quite difficult. (Sometimes you can get Q+DK for it, which is good enough, but this is quite rare.) And because Tenjiku has so many pieces of similar power (Lion Hawk, Great General, Vice General) the chances they get all traded away before a completely decisive imbalance occurs seems quite small. In Chu a much larger fraction of your total power is in that single piece.H.G.Muller (talk) 16:11, 9 February 2016 (UTC)

Well, according to the German Chu Shogi Association, you start with 161 pawns' worth of material (excluding the king), of which only 20 are in the lion (about an eighth). Once the lions are gone, we still have the next tier of rather powerful riders (the queen, dragon king, dragon horse, rook and bishop), that still can be very effective fighters. In contrast, in orthodox chess, we start with 39 pawns' worth of material, of which 9 are in the queen, almost a quarter. Furthermore, even in orthodox chess, trading your queen for anything but the opponents' queen is rare: yes, we hear of queen vs. 2 rooks and queen vs. 3 minor pieces, but how often do these occur in practice? Not very often.

Perhaps it would be possible to conserve the lions even without lion-trading rules if one were to attempt to beat back opposing lion attacks with loss of time by using one's dragon horses in the centre, strengthening this centre to drive the lion away with loss of time while developing pieces, and only using the lion later on whatever side you decide to attack on, where it will really help in the attack. Double sharp (talk) 12:27, 10 February 2016 (UTC)

This was essentially what HaChu tried to do, before I changed its evaluation to suppress it. The Horses are not very effective in harrassing the Lion, because they move diagonally, and the limited range between the Pawn lines (which usually step up to the Go-Between rank) then makes they cannot be displaced laterally efficiently. And the Horses then invariably get into trouble when a Copper sneaks out to attack them, and the Pawn wall blocks their retreat. Being in front of the Pawns is not a good location for Horses at all. So basically you are wasting more moves on your Horses than the opponent will on his Lion. In Chess harrassing en early-developed Queen works because you can do it with Knight, Bishop and Pawn moves that you would want to do anyway. But this is more like the Alhekine defense, where harrassing the Knight with Pawns just wrecks your own position. H.G.Muller (talk) 18:51, 10 February 2016 (UTC)

Well, I suppose I can't really comment before playing enough Chu, so I've been playing some games against HaChu 0.15 (yup, I downloaded it some time ago) on my computer. As befits a beginner to this game, I keep losing – but maybe I have hope, because I always ended up losing by hanging my lion, so maybe if I watched out more I'd be better. (Reminds me exactly of my own beginnings at FIDE chess hanging the queen ^_^.) Double sharp (talk) 06:55, 13 February 2016 (UTC)

Jumping generals and fire demon (from tenjiku)

How should their moves be symbolized in Betza 2.0? Double sharp (talk) 14:05, 9 February 2016 (UTC)

The Bishop General would be something like mBcBp0, which uses the quirky feature that a modifier on the range would overrule the default of the chaining operation underlying the exponentiation (normally m on the left operand and and f on the right operand, e.g. 'W0 = mW-fmW-fmW-...-fmcW). So in this case cBp0 expands to c(pB-fpB-fpB-...-fB) = pB-fpB-...-fcB, i.e. make 0 to arbitrary many Bishop moves in the same direction from platform to platform, and end with a capture.

The move of the Demon would be BsRK(K-aK)(K-aK-aK) = BsRKa3. That does not specify the burning, however. To do that requires a -xacK suffix on each move: explode and send shrapnell along any K move to capture. So (B-xacK)(sR-xacK)(Ka3-xacK). This does not describe the passive burning power, however, which Betza 2.0 at best could describe as a property of the other pieces. And this would be pretty complex, as every move of every piece would then have to test if it is not ending next to an FD. This can be done by having it explode destructively using xx, and let the fragments return from any square that does not contain an enemy FD to resurrect the piece. So with something like -xxamotp{!FD}K-bK as a suffix to all moves. This doesn't really clarify much, and a single sentence on how burning works would be much clearer. H.G.Muller (talk) 16:49, 9 February 2016 (UTC)

You know, one thing I'm uncomfortable about your "a" notation is the way it treats the lion. It seems to take that as the default case, so that "aK" implies that both steps can be either moves or captures (since neither is specified), which I am OK with, but also that intermediate steps can jump over other pieces, which is not how I'd imagine a doublemove to work. A piece that literally moved twice as a king would be possible to block, unlike the lion. So I mentally would prefer it if the lion were "ONADaK", specifying the jumps specially. (I've added the O to account for the case of the lion passing its turn while surrounded by pieces.)

I imagine it also implies that not all steps need to be taken, so that "aK" includes "K"; this is the way Betza's doublemove pieces work, where you do not have to exhaust their range. That's also the way it works in shogi variants, except for the thunderclap in ko shogi (IIRC). Double sharp (talk) 12:12, 10 February 2016 (UTC)

Well, since I first proposed this a modifier, my ideas on it have evolved too, and in the end the 'XBetza' notation used by XBoard (for the engines to reconfigure the GUI's move generator) did have slightly different defaults. The modality of non-final legs defaults to m rather than mc, and the directionality of continuation legs defaults to 'everything but straight back'. So aK would describe an area move without turn-pass ability (and without single K steps). To give it Lion power you would need mcpaK, meaning the first step can move, capture or hop. Because of the directonal default this still would not include igui or turn passing, so these would have to be added separately (as well as the single King move). This makes the Lion KmcpaKmcabK where the last two terms cannot be contracted, because one has p, the other hasn't. (Which reflects a true irregularity of the Lion piece.) As you remark this is equivalent to KNADcaKmcabK. That isn't nearly as compact as aK, but the latter would be defective no matter what, because it does not address the point that the Lion can jump over anything if it moves to another square, but cannot jump when returning to its own. I have come to like the notation with an explicit KNAD best, because it is the most intuitive. It explicitly gives all destinations that can be reached by direct leaps, for which paK might be more compact, but also much more obscure and contrived. The Lion power than has to be mentioned separately, but it deserves to be mentioned separately. And intuitively people tend to expect returning to the starting square to be a special case with its own rules, so mentioning that separately as mcabK to stress that you can return to the starting square for igui or turn passing is quite useful, rather than a waste of space. In Betza 2.0 it would read KNAD(cK-aK)(mK-bK). For the Lion Dog triple step I still don't have anything better than mcpafmcpavK (which, incidentally, also catches the single K steps though 2-out-1-in).

Making multi-leg specifications with a imply not all steps have to be taken seems a bad idea, because it really leaves no possibility for describing pieces where the step is mandatory. (E.g. a Checker, which is now fmFfcafmF, would then imply the Checker can also capture as a FIDE Pawn.) If a means a mandatory extra leg, you can always describe the case where the piece can also move without it by writing that explicitly. The a operator is also essential for defining lame leapers, like the Xiangqi Horse, because the 'old' Betza notation nN is ambiguous, and doesn't tell you whether the piece can be blocked on the W (Mao) squares or F squares ('Moa'), or both. In XBetza the Mao would be mafsW = afsW, a Wazir that (from an empty square) must continue outward at 45-degree (fs), while the Moa would be afsF, and a piece that can be blocked on both squares alarWaralW (cumbersome, because the chiral trajectory forces both handedness to be written seperately, but at least possible). None of that would work if the a was taken to mean that the Mao could also make a non-capturing Wazir move. 83.163.204.254 (talk) 18:37, 10 February 2016 (UTC)

Actually I think that turn-passing should be allowed by default, as doing otherwise introduces asymmetry and means that shooting has to be specified separately. If these are really doublemove pieces, then I think by default they should make two normal moves, so I would assume "mc" as default always. Then the lion becomes simply KNAD[aK], and a "strict-construction" doublemove king is simply K[aK]. (I've used the brackets because [KaK] is also conceivable, and means the same as [aK].) The hook mover is then maR, as the first step cannot be a capture: a doublemove rook is aR. The gryphon (ferz then outwards as rook) is FmFauR. I'm inventing "u" and "w" to mean outward and inward, so that I can distinguish a case like the Grande Acedrex unicorn (knight then forwards as bishop, IIRC), which is NmNafB. (This is why I wanted a to mean that you do not have to take all the steps.) In the special case when all steps are the same, this makes a without qualifications act exactly like my old d, and with all steps except the last one non-capturing like dd or Betza's t.

Note that even Jean-Louis Cazaux has now abandoned the idea that the Unicorn in Grande Acedrex moves like the diagram on his website. (He apologized to me in an e-mail that he had not had time to update the latter.) The current consensus is that a white Unicorn on e6 would move along (g5,h4,i3,j2,k1) or (f4,g3,h2,i1) on its backward moves, i.e. just like a black one moves on its forward moves. Which is how I made it move in Fairy-Max' Grande Acedrex implementation, through a NmpafsyafW definition. (I guess a more elegant notation would have been ygafsW.) Note that introducing a u and w for outward and inward is still as much ambiguous as just having f and b, so that you would also need alternatives for left and right in a relative rather than absolute sense. Apart from the fact that with the revision of the move of the Grande-Acedrex Unicorn the entire market for absolute directions on continuation legs seems to have evaporated, it still would be possible to specify such moves in XBetza or Betza 2.0 by relying on the absoluteness of the directionality in the first leg: fygafsWlygafrWrygaflWbmpafygabsW. That might look quite cumbersome, but for something you would never want to do that should not be considered a disadvantage. The whole philosophy behind the Betza system is that the commonly encountered cases of non-divergent fully symmetric pieces is simple, at the expense of making the description of highly asymmetric pieces cumbersome. From an intuitive POV one could argue that the various paths of the (now abandoned) Unicorn move were not geometrically congruent, and thus deserve to be mentioned separately. H.G.Muller (talk) 19:52, 16 February 2016 (UTC)

How would one then specify a shooting rook, that moves normally as rook, but captures by shooting? I used to use x as a modifier for "shooting", so that this would be mRcxR. [Because I think "mxR" would let it shoot at empty squares.] I don't think one can do better with a as then it becomes difficult to force it to step the same number of steps backward. A piece that moves without capturing as ferz and then can shoot pieces like a rook (but not move that way) is mFacxR.

In Betza 2.0 I introduced the e modifier for this, requiring the length of the sliding leg to which it applies to be equal to that of the preceding sliding leg. This would make rifle capture cR-ebR or cB-ebB. But it would also allow definition of pieces like 'Sissa' (mR-esR, a restricted hook mover) or 'Equi-hopper' (pQ-efQ). Sadly I did not define anything like that for XBetza, though. And I requisitioned the e there as convenience modifier for e.p. capture, which was pretty essential for defining non-standard Pawns (as several of Fairy-Max' pre-configured variants had to do). I suppose I could overload the i modifier for this (abbreviating 'iso'), as the current meaning as a virgin-only move only makes sense on a first leg, while the 'iso' meaning only makes sense on continuation legs. So rifle capture would then become caibR.

The case of a fire-demon burn is something like Betza's distinction between immediate and deferred effects. When a fire demon moves next to an enemy piece, it is immediately burnt (shot). But when a heavenly tetrarch moves next to an enemy piece, the piece is not yet captured. The capture (shooting) does not happen with the heavenly tetrarch's move, but must use up another move, though in neither case does the piece move.

I am tempted to say therefore that some things are better written in text than in Betza notation. Double sharp (talk) 15:47, 14 February 2016 (UTC)

I think I have a reason for why I would like a to include all steps: that's the way Betza notation usually works. R3 means "up to three steps", not "exactly three steps". The gryphon was classically t[FR], but he let it stop at the F squares, so it appears that zero squares is included for standard pieces, but is usually suppressed as it doesn't change the position in that move. (Doublemove pieces can do this as they take multiple steps. Otherwise they can't shoot.) Similarly, since p allows you to jump over one hurdle, p2 should let you go over at most two, and pp any other (just like W, W2, and WW). In fact, I'd suggest that to specify exactly two, you could write R2!. For a piece that must take the second step, you could write a!K. I might similarly use a3K for a piece that takes up to three steps, and a3!K for exactly three.

Betza 2.0 uses the method (stolen from Bex notation) to prefix the range with a redundant 0 to indicate exactly that number of steps, like R02. When I was designing Betza 2.0I did consider to use the t on a non-final leg in a path to indicate it could optionally terminate there. That would give F-tR for the Griffon. But this would not save a big deal compared to F(F-R). Note that the Griffon can also be written as mpW-sR, adding a 'virtual visit' (mp) to the W squares, making the F squares automatically a possible termination point of the Rook leg (ygasW in XBetza). In Betza 2.0 the notation with a numeric range following a modifier is used to alter the chaining defaults away from (m,f). So Ka3 there would expand to what you mean above by a3K: K(mK-aK)(mK-maK-aK). (Note, however, that the similarity is partly coincidental, the a in Betza 2.0 not implying chaining but meaning 'all directions'.) H.G.Muller (talk) 19:48, 16 February 2016 (UTC)

The fire demon can move as BrlR, or it can make up to three steps, a capture stops the move. I would write this as mKa3K, so that only the final step can be a capture. Finally it has an immediate shooting-king component, that is resolved before seeing the similar capabilities of an enemy adjacent fire demon. This is then "immediate xK". (An exception would be needed under Colin Adams' rule where both fire demons burn adjacent enemies.) Thus anything that ends next to an enemy fire demon is removed from the game instantly.

The tetrarch on the other hand is an interesting anticipation of Betza's anti-rooks. A R3 can go only to the 3 nearest squares a normal R could reach; for the R3, it is the 3 farthest. But I would like to redefine it, perhaps more intuitively (but also to allow you to change the board size), so that the R3 can go up to three squares and the R3 at least three. For ASCII I would use inequalities like R(>2). Thus the tetrarchs may be written B(>1)fbR(>1)rlR(1<n<4)xK. The picket from Tamerlane chess is B(>1), and the giraffe from the same game is Fa!R(>2) IIRC.

As for the lame leapers, Betza's defaults seem to be orthogonally as far as possible before switching to diagonal. Thus "nN" is the mao if you don't specify it further. Fully specified, it is Wa!uF. The moa is Fa!uW. If you can block it both ways, then it is Wa!uF AND Fa!uW, since it can choose either path but both must be open. (If only one needs to be open it is Wa!uF OR Fa!uW. I think the XOR version is a most interesting piece!)

Now the question of the HF and SE (and LD in the ja.wp rules) is how to confine "a" to a single dimension. I could postulate a 3D doublemove piece, that for its first step moved as king, and so too for the second step, except that the second step had to be a 2D move that kept it on the same level. Given that Betza thinks a White rR on h1 has no moves, but a Black rR on h8 does, though, his modifiers are obviously relative. So "fb" as now seems fine. Where it doesn't work is the ko shogi piece that does this with knight moves, as it wants two different forward steps but only the starting point as a backward step. But just NxNaffN works. Double sharp (talk) 00:14, 15 February 2016 (UTC)

Well, as you see it is not difficult to design Betza-like move-description systems, and there would be many ways in which it could be done, each equally satisfactory on its own merits. Betza wrote somewhere he intentionally avoided the use of punctuation characters, because their availability might depend on the platform or locale. I am not sure this concern still applies these days, and he violated the principle himself when he introduced the t[AB] notation. In XBetza I selected the m + all-but-back defaults, because these were what you most-often wanted, and thus would lead to the most compact notation in the common cases. Note that lion power is quite rare amongst Chess variants, and that the most common application of the a modifier is to precisely define lame leapers like the Moa. Betza's ad-hoc disambiguation of the n modifier by letting orthogonal prevail over diagonal is of no help when you want to describe the Moa or multi-path lameness. Then there are bent sliders or hoppers and hook movers, which are stil more common than lion power. So the notation was optimized for those. The notation Fa!uW you suggest for Moa is based on a system more similar to Betza 2.0, where the specification of a path can use multiple atoms. Betza 2.0 uses a hyphen for chaining different atoms, which is probably optically clearer. I don't particularly like XBetza for explanatory purposes, (Betza 2.0 often seems much clearer), but it was easier to write a parser for it, and it does have the expressive power to do almost anything that was required at the time (i.e. all pieces in all variants supported by Fairy-Max, Shokidoki or HaChu). This is why I picked it for use in XBoard's engine-GUI communication. And that basically was an irreversible decision; backward compatibility is sort of sacred in XBoard. So XBetza can be extended, but only in a backward-compatible way. H.G.Muller (talk) 19:48, 16 February 2016 (UTC)

Ko shogi

Do you intend to do anything with ko shogi? It seems like it may give tenjiku a run for the money in weirdness. Double sharp (talk) 11:42, 16 February 2016 (UTC)

Tenjiku engine

A while ago at Talk:Maka dai dai shogi I suggested turning the FAD from the Clobberers into an maF, which you confirmed to be balanced, and you suggested turning the R4 from the Rookies into a R3. I was wondering if you have since tested the latter suggestion, or if you have any suggestions for weakening the Nutters. ^_^ (I tend to find the Nutters to be rather ugly and asymmetrical pieces anyway – at least the FIDE pawn simply has no backward moves, but not different backward moves – but because they are one of Betza's original armies, I feel somewhat obligated to deal with them too.) Double sharp (talk) 04:21, 25 August 2017 (UTC)

No, I haven't done much testing with Fairy-Max lately, (except that I hacked it to make a dedicated version for playing the ICC variant TwoKings, to help settling the question whether the switching royalty made the second King stronger or weaker than a normal Commoner). This because I (and in particular my computer) was too busy running my new Tenjiku Shogi engine, for playing in the Tenjiku correspondence championship on gamerz.net, and developing opening theory for it. H.G.Muller (talk) 07:37, 26 August 2017 (UTC)

Even more interesting then! ^_^ I wonder if we will have this available in your wonderful WinBoard Shogi-Variants package at some point? Double sharp (talk) 15:05, 26 August 2017 (UTC)

That is indeed the plan. Originally I wanted HaChu to play all large variants up to Tai, and indeed a lot of the code for Tenjiku was already working. I never got to implementing the planned incremental update of the attack map from which HaChu generates its captures, however, and for variants larger than Chu this really starts to hurt badly. (It now generates the attack map from scratch on every move, defeating the purpose.) And the attack-map format HaChu was using (keeping a count of the number of attackers from each of the 8 principal directions) would still require the move generator to search for jumping generals in those directions (which could be hidden behind many pieces). In the mean time I had designed a better way to handle jumping generals, so finishing HaChu's attack-map handling made little sense, and I wrote a new engine from scratch based on the new ideas. I just have to decide whether I will now merge this with HaChu, to make a HaChu 2.0 that will play all variants, or whether Tenjiku is so special that I keep it as separate dedicate engine for it ('Inferno'). I also have to decide about what rules I will use in the released version. In particular vertical vs sideway Fire Demon, instant burning on WB promotion, and whether I will support jump-capture of royals (as in the '21st-century Tenjiku' played on gamerz.net). I think I am close to proving (through analysis of opening theory) that the first-move advantage in 21st-century Tenjiku is as least as much as that which led to abolition of the TSA rules (where it was only a Queen). I still have to address the question whether using a sideway Demon can cure that, or whether jump-capturing of royals is an untenable idea (and better not supported). Problem is that this jump-checking is considered the main attraction of the game to the community of 21st-century Tenjiku players, which sets it apart from other large chess variants, and makes it worth playing. I am also getting worried about the repetition ban: if the 1703 description of Chu doesn't mention it, and it comes from the 1928 rule description by Okazaki, the latter might very well have introduced it, as a contamination by modern Shogi. In drop games repetitions are virtually the only way to cause a draw, and it does make sense to somehow rule out those remaining draws. But without drops the large majority of draws would be due to insufficient material to force mate in an end-game. Forbidding repetitions could in theory artificially decide those, but except for K vs K this would require to play on for thousands, if not millions of moves, and it is unlikely any chess player would like to force such an artificial decision anyway. So if the rule was not for that, then why bother with the tiny fraction of draws caused by perpetuals, and make a special rule for those? Even for end-games like Queen vs two Tigers a ban on perpetual checking by the Queen is of no practical help for allowing the Tigers to force a win: the Queen can check hundreds of times before it has to make a first repetition, and it can start that anew for every move the Tigers make. And if a repetition ban is a recent invention for Chu, Tenjiku (and perhaps other large variants) will almost certainly never have had such a ban. H.G.Muller (talk) 19:06, 26 August 2017 (UTC)

I think it is fairly well-established that the Fire Demon must have had the sideways move. Firstly, that is what Japanese Wikipedia ascribes directly to the SZ and SSZ: "『象戯図式』および『諸象戯図式』では、「水牛のように動く」となっており、横と斜めに何マスでも進めることが示唆される。" That is, "The SZ and SSZ have it keep the move of the water buffalo, so that it ranges instead horizontally or diagonally." Secondly, Peter Banaschak, who apparently had access to the SZ, translated the relevant portion in Colin Adams' book as follows: "Fire Demon (two steps back and forth, ranges to the left and right and in the diagonals) [moves] like the Water Buffalo, furthermore [makes] three King-like steps. Enemy pieces on the eight adjacent squares perish by burning." There is not much room for argument there, it seems.

True, but things cannot always be decided by arguments. Problem is that the rules used by people that actually do play deviate from the historic rules in many respects, and that these people apparently like those rules. A strong engine can influence them, but only if it will not be ignored. Nothing will be achieved by alienating the primary user base, and the current '21st-century-Tenjiku' champion openly declared he would not be interested in a game where the Demons move sideways. I think I can get them to adopt the correct FE and HT move in the 21st-century variant with only a little bit of grumbling, but I am not so sure about the other issues (Demon direction, jump checking and +WB burning). A too-large first-move advantage would no doubt be a powerful argument to change their current rules, but they would have to be convinced of that through thorough analysis. (When I would publish a winning line for sente, I guess they would have no choice but to change something to keep playing it interesting. Although some already say that they could just keep the current rules, and adopt the 'pie rule' to make it fair.) H.G.Muller (talk) 19:38, 27 August 2017 (UTC)

I find the idea of using a "pie rule" rather odd: in practice that would just result in a restricted first move, as no one would allow White to start with the instant win. So if he starts with it, Black would demand to switch sides, and win instead. But if White makes a passive move elsewhere, which would then be accepted, then I don't see how he can prevent Black from starting likewise with the instant win. Is there some first move that stops the other side from doing this? Double sharp (talk) 23:48, 27 August 2017 (UTC)

I've since come round to your interpretation that upon Water Buffalo promotion to Fire Demon, all enemy pieces are burnt immediately. Because moving next to an enemy Fire Demon is always a suicide move, this would be the only case where an enemy piece is unharmed by staying next to a Fire Demon, which seems fairly odd.

Indeed. As a compromise I had them change the PBeM server so that at least the passive burn on the next turn will vaporize those pieces when the +WB survives. But I don't like it; it looks very unnatural. With an immediate burn the Buffalo does become almost as dangerous as a Demon, however ('nuclear-tipped'), as you can run it into the Pawn line, burning three of the generally expensive pieces directly behind that line. So this rule has a huge impact on the game. It would have to be checked whether this will not unacceptably alter the first-move advantage. It could make up for the less aggressive (sideway) Demon in terms of excitement, though. So perhaps that would make it an acceptable 'package deal' for the current players, if it turns out that it solves the problem of the first-move advantage. But it would take a tremendous amount of analysis effort to show this convincingly. And although my engine currently is usable, it is far from being finished. I now sometimes have to analyze more than 24 hours to recognize certain threats, and perhaps future versions with better move ordering could see the same in under an hour. So the prudent course of action is to first improve the engine, then analyze openings for various rule combinations, and then decide how strong our case is to get the required rule alterations (from current practice) accepted. I can then release the engine supporting these rules (in addition to the historic rules) through options settings. H.G.Muller (talk) 19:38, 27 August 2017 (UTC)

Well, it is really a pity that there are no historical Tenjiku problems! Indeed, I think the next-best bet in the absence of an iron-clad instant win is to offer options for all the controversial matters: Demon direction, Water Buffalo promotion, and jump-checking. (Do you offer options for the Chu promotion, baring, and repetition rules?) But in the meantime improving the engine takes the highest priority.

This behaviour of the Water Buffalo actually seems like how most beginners to Chu mistakenly think of the Kirin. The difference is only that the Lion's igui is a deferred effect and takes up a move by itself; while the Demon's is an immediate effect. But what a difference! And how different both are from the promotion of the FIDE Pawn to an immediate game-decider, because it can only promote on a limited number of squares; and since its movement and capturing powers differ, this opens the possibility of the Prokeš-manoeuvre!

BTW, is there some place where these discussions on the rules you have had with the current players can be read online? It would be very interesting to see. Double sharp (talk) 23:48, 27 August 2017 (UTC)

I have difficulty imagining why jump-capturing might attract anyone to Tenjiku specifically when Xiangqi already has a form of that. Rather it was the Fire Demon that would seem to be the main attraction, just as the Lion is for Chu. To keep these games interesting you need to have extremely powerful "superpieces" at the top of the hierarchy, and a reasonable amount of lesser pieces below to keep the balance, but they had better have easy-to-remember and rather symmetric moves; already the variety of limited Kings and Queens that make up the bulk of the piece types in Dai, Dai Dai, Maka Dai Dai, and Tai are really annoying to remember and not all that strong that you care very much, despite the latter three atoning for it by including Hook Movers and Emperors. For this reason I find Ko Shogi more intriguing than any of these four: I wonder if you intend to make an engine for that too?

Well, I did not express it accurately. Of course the FD is the main attraction. But the rules they play it by, as implemented in the PBeM server at gamerz.net, lead to a game where the typical tactics is to blow up the opponent's generals that could block yours at the expense of a Demon, to threaten a smothered mate with your own jumping generals. Which the opponent then can only prevent by sacrificing his Demon for fewer generals (if at all). It has attracted people who do like this interplay between Demons and jumping generals, which would disappear when the generals cannot jump-check. As for Ko: this has been largely below my horizon, because Steve Evan's ShogiVar did not support it. It could be interesting. But I will do Macadamia / Maka Dai Dai first. (I have already started designing efficient update algorithms for hook moves.) H.G.Muller (talk) 19:38, 27 August 2017 (UTC)

I would suspect that the existence of such a forcing opening is already a sign that something is wrong, but I'm probably underestimating how attached the current players must be to the current rules: with that sort of attachment, only a demonstration of an instant win will help matters. Double sharp (talk) 23:48, 27 August 2017 (UTC)

Likewise, I can only agree with you regarding repetition. Double sharp (talk) 15:10, 27 August 2017 (UTC)

BTW, has your work on the engines for Chu and Tenjiku resulted in more accurate piece values, or are you still using the values from the website of the German Chu Shogi Association? And what are you using for the other variants, which except Dai-Dai have no values listed on their website? Double sharp (talk) 06:53, 28 August 2017 (UTC)

I have been using the values of the tenjiku-shogi.net website (which seem to be Dave Harper's). They prove grossly inadequate. For one, most '21st-centurey Tenjiku' games played by people did not last more than 10 moves. So they all apply to the game phase where the Kings are smothered, and you try to quickly checkmate it. In this game phase the evaluation is dominated by a King-safety term that reflects smothering of the King, presence of generals that could checkmate it, presence of potential checking squares out of Demon range (e.g. moving the FE aside to make room for a GG is often a killing move), and centralization of a Demon. Trying to attribute that to contributions of individual pieces completely distorts the values. In the middle-game a Demon is 'priceless', (as I already guessed in my own game against kokosz, but which I had not programmed into the engine), and was demonstrated in game 1903 of kokosz vs bot_hachu. The engine traded a Demon here for a collection of high-valued pieces (FE, Q, Ln, LH), but with no Demons against one it lost. It seems a Demon would win an end-game against almost any amount of other material, provided the King has some protection against checking. There just is no defense against it. H.G.Muller (talk) 09:38, 28 August 2017 (UTC)

Of course, instant-win tactics and unparriable threats grossly distort piece values. (This must make pieces with deep forward leaps of 3 or more very dangerous in a FIDE-like variant; they cross the board very rapidly and can attack from out of reach of the Pawn line. Which makes me wonder how much of the Rookies' too-high strength comes from the HFD, but that's another topic.) This may unfortunately make it completely impossible to calculate the value of horrible things like a doublemove Queen, because beyond a certain level of strength the value of the piece rises further than mobility would imply because of the unparriable threats of mate or material gain it results in!

From what you say, I am slightly worried that the Fire Demon's power is already sufficient to cause this problem of pricelessness, and then all that removing its forwardness would do is to make it just take a little bit longer to do the same damage. Is it even possible to keep out a Fire Demon when you do not have one to counterattack with? That's the problem with giving ranging movements to powerful pieces at close range: a Lion can be held at bay, but a Lion Hawk quickly uses the Bishop move to run to any weak spot in the defence you have, and no dout a Free Eagle would be similarly deadly. Like I said a while ago, I wonder what this would mean for mixed pieces like the princess or empress on a 16×16 board: from afar, they look like a bishop and a rook respectively and their knight move has no influence, but their bishop or rook move is immensely useful for getting them to a position where their knight move is at its deadliest. Double sharp (talk) 10:34, 28 August 2017 (UTC)

Indeed, the power of the Demon is worrisome. The threats of smothered mates can at least force the opponent to sacrifice a Demon back, after you blew up his protective generals shield with one. (This often works by capturing into his Pawn line in front of your GG, (his VG), forcing him to recapture with a Demon to get a burn-cover on your VG square where you would otherwise checkmate him with your GG on the next move. Or by forcing him to parry an RG check with the RG in front of his Demon, exposing the latter to your other RG.) Without that it seems difficult to force a Demon trade, and sacrificing a Demon for other material seems a fatal mistake. Except perhaps that the main stategic motif of the game becomes to promote a Water Buffalo, possibly sacrificing a Demon to achieve that. It should become much easier to promote a WB when it burns immediately the pieces that might be able to recapture it. (Btw, the HFD did not prove very problematic. I guess this is because it only has a single forward range-3 move, so it cannot fork anything on the opponent's backrank. And the point of possible attack is easy to predict, and his fF threats can be countered with a single Pawn. But with pieces like Bison (CZ), or even Grasshopper, it was indeed a disaster. This can be cured by moving all Pawns up to 3rd rank, like in Makruk, though.) H.G.Muller (talk) 11:16, 28 August 2017 (UTC)

And this degree of forcing isn't enough to convince people that something is the matter with their current rules?! ^_-☆ I likewise don't see how you can survive without a Demon if your opponent has one, and it seems as though no amount of material would suffice as compensation for a Demon except another Demon. This sheer non-linearity of the scale seems surprising: I have to wonder how much normal material would be needed to overpower a Demon, if it can be done at all.

Well, the forcing doesn't seem to lead to a bad game. It helps to trade the Demons out of the game, the player initiating the trade gaining some material in the process, and thus promote aggressive play. It is just that the first-move advantage seems too large under '21st-century' rules. But perhaps switching to a sideway Demon would already cure that. Pieces like Lion and Lion Hawk are probably less valuable because most pieces running the show are 'Lion proof'. H.G.Muller (talk) 13:07, 28 August 2017 (UTC)

But how much extra material? With the interpretations from Colin Adams' book the advantage is a Queen, which seems to be decisive already with good play. I agree that the Lion is probably less valuable, but perhaps the Lion Hawk would be significantly more valuable than the individual Lion and Bishop, because the moment a gap is left in the defence, the Lion takes a long time to get there, but the Lion Hawk can get there immediately. Now I wonder what good Tenjiku strategy looks like on how to use the lesser pieces: with Chu it is more obvious, using many steppers in the King's castle, using the Side Movers to guard the promotion zone, etc., but with Tenjiku it is not clear how to use some of those sand grains in the asteroid storm (what to do with things like the Ferocious Leopards, far too far out on the wings to use in castle-building?). Double sharp (talk) 15:24, 28 August 2017 (UTC)

Indeed, but Colin's book is for TSA rules. I was surprised how valuable a tempo is even after the Demons are gone; in my game against kokosz (1885) there was just no time to make a single stepper move without losing material. When I left the opening line prepared with the engine's help, I wasted two tempos by launching an ineffective attack, and that basically cost me the game. H.G.Muller (talk) 15:55, 28 August 2017 (UTC)

Which basically confirms my suspicion that the first-move advantage is too great under "21st-century" rules, if you cannot have a single spare tempo to do something other than avoid immediate disaster. ^_^ This makes me hopeful that with enough depth there should be a clear instant win available. Double sharp (talk) 16:13, 28 August 2017 (UTC)

But at that time all jumping generals had been traded (except one BG of mine, on the other shade than his King) as well as the FD, so the rules had basically converged. H.G.Muller (talk) 18:17, 28 August 2017 (UTC)

But one side still had a material advantage out of the opening, thanks to the ruleset, no? If that was not the case, then I think that, as fun as it is to take out eight pieces at once, Tenjiku might well have the insoluble problem where both sides are skating on the precipice all the time, and one false move leads to retribution with extreme prejudice; and with this much tactics, and with this large a board, it may well be too complicated for most people to play without blundering every other move. (One of these factors would be okay, but surely not both.) So the engine would be invaluable. ^_^ So perhaps it is not just the ratio of force to space that matters, as Betza suggested, but also that beyond a certain amount of force or space it becomes really hard to hold everything in mind at once. Double sharp (talk) 23:25, 28 August 2017 (UTC)

True about the HFD; like you I suspect that this is because the H cannot fork forwards. Otherwise disaster would strike instantly. I suspect that without this (as can be tested with a Makruk-like setup, as you say), the Bison would be a lot less valuable on an 8×8 board because its stride is so long and awkward; on 12×12 it should be a reasonable piece. Double sharp (talk) 11:32, 28 August 2017 (UTC)

I once did test the Falcon from Falcon Chess, a lame multi-path version of the Bison, and it came out close to a Rook. This would be low for a 16-leaper (normally leapers with 12 targets are of Rook class), but the lameness was probably more responsible for the value deficit than the large range. H.G.Muller (talk) 13:07, 28 August 2017 (UTC)

There is something very odd about calling a lame piece a Falcon; surely one would expect a bird to fly over obstacles. But that is interesting: I would expect a normal Bison to be worth about halfway between a Rook and a Queen if I ignore the range for a moment, and if I mentally remove the lameness this actually seems to be correct even on the 10×8 board of Falcon Chess. Perhaps part of it is the wideness of the board; maybe another part of it is that the combination of Camel and Zebra allows more flexibility than either alone?! Double sharp (talk) 15:17, 28 August 2017 (UTC)

It would indeed have been more fitting to name it after some snake, as it tends to slither between the other pieces. The weakness of Camel or Zebra on 8x8 becomes critical in the end-game, where they have very low manoeuvrability because most moves fall off board (or will so on the next move, after you chase it away from a central square), and there are not enough pieces to protect it. Then it will often be lost without compensation. I suppose the Falcon suffers much less from that, because when the board population thins, the lameness gets less of a problem. And because of its 16 move targets the Bison still has a reasonable 2-move mobility even on 8x8. But that the board is 10x8 will certainly help; it can be shuttled between two files (d and g) where it doesn't hit the side edges yet. H.G.Muller (talk) 15:55, 28 August 2017 (UTC)

Last year my (still private) Tenjiku engine won the yearly (Modern) Tenjiku correspondence championship. But it had sente against Kokosz, the only other serious contender. This year it will be reversed, and the result will be more telling. From analysing opening, and my own games against Kokosz, I acquired the following insights: (1) The statement that a Demon is priceless needs to be qualified a bit. If player A has a Great General, and player B not, and also no Rook Generals, player B's King, even if no longer smothered so that mate is not possible, can still be checked from the fromt from safely inside player A's camp whenever A so desires. This makes any Demon of B potential victim of a fork, either diagonally from within A's camp, or from the side (in which direction it doesn't slide according to modern rules). To do the latter the GG can break cover, sneaking out of its camp by 'ladder checking' until it reaches the Demon rank. So the value of a GG depends very much on the presence of enemy Demons; without those it is just a strong (unblockable) Queen, but otherwise it could be worth a Demon, either by actually capturing one, or by preventing they can be usefully deployed. If the opponent has two Demons, these can also be forked, even with TSA rules (where jump-capture of royals is forbidden, but Demons are vulnarable). So perhaps the loss of a Demon for a GG in the opening is not as fatal as was once thought when dismissing TSA rules. If there are still RG on the board they can be used to protect the King from frontal GG checks, but then it depends on the balance of RG whether the defending RG can be easily traded away to get an 'unchained GG'. (2) Betza's 'leveling effect' is very large in Tenjiku. A Bishop or even Dragon Horse are obviously weaker pieces than all these 'nearly Queen' sliders. But you don't have to worry too much about losing a nearly-Queen to a B or DH, as it is almost impossible to prevent it. There are so many nearly-Queens, that it is almost impossible to shield them all from B or DH forks and skewers. Also, since tempo is worth so much, trading a developed DH for a nearly-Queen still in the start position might not be such a good deal. I lost my game against Kokosz last year because a Demon trade destroyed his Pawn rank and allowed his light sliders to easily develop, and 'poke me in the eye', and I spent all my moves cowering my valuable pieces away from them. It would be better just to accept he traded them for nominally somewhat stronger pieces, as my time to do the reverse would almost surely come when I manage to develop my Bishops and kin. So nominally weak sliders get a large value bonus as long as there are still enough more-valuable opponent pieces to trade them for by forking and skewering. (3) After analyzing more opening lines, I still haven't managed to find one that forces an obviously winning advatage for sente. Surprisingly enough advantages that at first seem huge always tend to collapse when I follow the line deeper. Of course the evaluation of the program is still very unreliabl, but still...

---

If the repetition rule is not present in the 1703 document, then that must surely mean that it is not in any of the historical variants, and that repetitions when no one is trying to make any progress towards capturing royalty should always be draws, right? And if so, should we update the articles to remove this anachronistic rule? Perhaps by saying that there is no evidence for a rule against repetition, and for definiteness it can be assumed that it results in a draw? (Except of course for chu shogi, where we of course must present the historical rules together with those used today by the Japanese Chu Shogi Association?) Double sharp (talk) 11:22, 29 August 2017 (UTC)

Far-rank promotion in shogi variants

BTW, I just checked the Japanese Wikipedia article on Chu, and it appears to say that promotion of Pawns at the twelfth rank is not forced, and you can make it end up as dead wood just like the Lance. ("後ろに下がる事のできない香車と歩兵が敵陣1段目に不成で進んだ場合でも、不成を選択できる。 ただし、そうなった場合は、その香車と歩兵は死に駒（行かずの駒）として扱う。") I suppose this may still be of value when both Lions are on the far rank and you need that Pawn to not stand as a "capture bridge", and this seems more consistent with the non-idiot-proof promotion rule for Lances. BTW, the Chushogi Renmei website agrees: "歩兵・香車が敵陣の最後段まで進んでなおかつ不成の場合は、その場で「死に駒」として扱う。（今の小将棋だと禁じ手ですが、中将棋では許されています。）" So I will change the articles on Chu, Dai, and Tenjiku: the rule of Pawns is that you get a second chance to promote at the far rank on a non-capture, and not that you are forced to promote. Double sharp (talk) 15:10, 27 August 2017 (UTC)

I guess this escaped me because of the awful quality of the Google translation I thought the 'dead wood clause' only applied to the Lance. It does make sense that a Pawn would also have this option if it exists for the Lance, though. H.G.Muller (talk) 07:19, 28 August 2017 (UTC)

Previously, I imagined constructing these "complete rule reconstructions" that the rule for Pawns in Chu Shogi could be generalised to non-upward-compatible promoters in general, like the Knight in Dai Shogi and the Knight and Iron General in Tenjiku Shogi. This may be overstretching the historical sources, though. As you say, the Knight and Iron in Tenjiku are "sand grains in an asteroid storm", and have no hope of reaching the far rank. I am not sure about this: do you have an opinion? Double sharp (talk) 15:31, 27 August 2017 (UTC)

---

Also, I asked you a few questions on Tenjiku promotion of the Lion at Talk:Tenjiku shogi#promoting lion movers on in-out moves (specifically whether igui into and back out of the zone should allow promotion, and whether promotion can be obtained by zigzagging into and out of the zone when both the Lion and its single victim are just outside it). Double sharp (talk) 11:00, 29 August 2017 (UTC)

rebalancing CWDA

Well, after two years, I hope you don't mind that I bring this up again, in the hope of allowing myself to play something other than FIDEs vs weakened Clobberers! (I just dislike playing with armies that aren't really equal in theory even if the initial familiarity with the FIDEs will boost their value for humans, because that must eventually run out.)

I've been rereading the comments to the CWDA page on CVP (including many of your wonderful ones about piece values and tablebases); if I understand them correctly, the Nutters beat FIDE by about 1.5 pawns and the Rookies are stronger still. The most natural way to weaken the Rookies would probably be changing the R4 to a R3 as you suggested; after all, Betza himself vacillated between R3 and R5 before settling on R4. How much are short rooks (W, R2, R3, R4, R5, R6) worth on 8×8 compared to the rook, though? Part of the point of the R4 was that it was supposed to be thought of as an overstrength minor piece more than a weak rook (and the value you assigned it in Fairy-Max according to one of your CVP comments was 400 centipawns compared to the 325 of a knight and the 500 of a rook). Would an R3 then overturn this mechanic by being too close in value to a B or N? (Because then we already have a slow-to-develop minor in the WD; the Rookies almost have to violate the beginner's maxim and develop the rook-class HFD first, because the threats it creates win the tempi needed to develop the slower WD and short rook.)

I also understand that individually each Nutter piece is about equal to its FIDE replacement, except for the overstrength charging knight, because of the great synergy between K and N moves. (Isn't the KN compound worth almost a queen on 8×8?) Would it suffice to weaken it by making it a fhNbK with only seven targets? It seems to me more difficult to balance this army in a way that remains pretty true to how it works. (Although I confess I do not really find the asymmetry of the Nutters aesthetically pleasing.) I still prefer weakening the Clobberers by turning FAD into FmaF rather than (as suggested in the CVP comments) turning BD into BnD, because it is close to something Betza also suggested (i.e. turning the FAD into a doublemove ferz); that's why I also like weakening the Rookies with the short rook. But I don't remember Betza saying anything about possible alternatives that he considered for the Nutters. Double sharp (talk) 04:29, 23 March 2019 (UTC)

P.S. I notice that summing the Fairy-Max piece values you posted seems to make the original Clobberers look the strongest of all, because both the BD and the FAD are rook-class and this isn't compensated much by the NB's minuscule difference from a queen. Since the order seems to be Rookies > Nutters > Clobberers > FIDEs, this seems to need an explanation. Are the Clobberers significantly less valuable without their pair bonuses? And are there pair bonuses between different piece pairs, i.e. is there still any pair bonus if you only have one BD and one FAD, but they are on opposite colours?) I realise that you can't always simply sum values like that (e.g. 3 queens vs 7 knights on 8×8), but if you can't for some case I'd expect there to be some reason specific to that case that explains why it doesn't work. Double sharp (talk) 05:40, 23 March 2019 (UTC)

I haven't done much in the area of piece values lately. Greg Strong has been active in this area for CwDA, and posted on CVP about it, but he also seems to have frozen this before a final conclusion was reached. I did all the limited-range Rooks once, but it was so long ago that I don't remember the result, or even where I wrote it down. :-( Color-binding is indeed the most-likely suspect for why the Clobberers as a whole under-perform. Normally mating potential doesn't contribute very much to piece value, but perhaps if too many pieces do not have mating potential it starts to hurt in a non-linear way. Also beware that Fairy-Max does not support under-promotion, and that I implemented CwDA such that each army always promotes to its own super-piece (rather than the opponent's super-piece when that is stronger, as the official rules seem to allow). The Clobberers have the weakest of all super-pieces. Not by much, but enough so they cannot hold the draw in the face of a further disadvantage of a minor. (BN vs Q+B or Q+N is a general loss.) Perhaps the number of games where both sides re-acquire a super-piece through promotion is large enough to express this in the value of the Pawn.

Another concern is that Fairy-Max isn't really suitable for measuring pair bonuses, as these are not in its evaluation. I have a derivative of it that I called Pair-o-Max which at least can attach a bonus to each pair of identical pieces (in addition to understanding lack of mating potential). But it still would not do cross-bonuses like those that should be expected between FAD and BD. What I did establish is that 4 Bishops count as 2 pairs, as far as bonuses is concerned, and that the 3rd Bishop has the same value as a lone Bishop. It would be nice to redo all this with a somewhat more advanced engine. At the moment I am very much occupied with work on the 'Jocly' web-interface, though, where I am working on implementing Chu and Tenjiku. (See http://hgm.nubati.net/jocly .) I entered Jocly as a bot in the Tenjiku correspondence championship, which started March 1, so that its clock is already running. More on Tenjiku eval in the appropriate section. I also did 'Scirocco' for Jocly, which is kind of a mix between Courier Chess, Tamerlane Chess and Chu Shogi that I really like, but which no on else seems to have the slightest affection for.

Well, Betza wrote on R3 "Experience disagrees with calculation. Calculation shows that the R3 should be a bit stronger than a Bishop, but experience indicates that it may be a bit weaker; in both cases, it is fairly close." If they are that close in value, then as you have mentioned in several CVP comments their values should be pushed towards each other just like B and N are. That suggests that some of the fun of having R4 instead may be lost if we change it to R3, because R4 vs B would be an interesting "minor exchange" that easily becomes practical for positional reasons. Isn't the fun of CwDA that you are always in a situation of material imbalance, at least until you trade down to a pawn endgame? ^_^ Still, I think the mating potential of R3 ought to keep some interest in this exchange.

Mating potential is surely important; there's no question things would be very different in FIDE chess if KR vs K was not a win. Before writing this I was going to talk about a thought experiment in which no single piece had mating potential, but then I remembered that your Team-Mate Chess has a low drawing percentage despite no single piece having mating potential, so there must be some counterbalancing factors. Maybe the situation seems more dire in FIDE because the rook is slow to develop in a way that the Team-Mate Chess pieces are not, though, so that rook endgames (which can potentially simplify to KR vs K) occur disproportionately often. This suggests to me that the Rookies may also be winning a lot due to their endgame advantage; their minor pieces also have mating potential, so their values are not depressed as significantly as those of the B or the N as the last pawns disappear. So they have a funny situation; FIDE has easily developed minors and a slow rook-class piece, whereas the Rookies have the opposite going on (although the RN is of course about as fast to develop as the Q, it is not a good idea to do it too early). The Clobberers of course have everything pretty easy to develop, whereas the Nutters have a FIDE-like situation when only the "rooks" are slow.

Perhaps what you mention about reacquiring super-pieces through promotion is what Betza was thinking of when he allowed promotion to the other colour's pieces; the only situation when a promotion can happen but is not going to be immediately decisive is a pawn race (KQ vs KP counts too, I guess, due to the need to deal with promotion threats), where a pawn endgame turns into a queen endgame. In this case the Nutters would definitely be well-advised to promote to the other side's queen replacement!

(Inter Scriptum: Vladimir Roytman said something to this effect during an email game of FF vs CC with Betza: the quote is below.)

I think most of the pawn's value comes from control of squares, pawn structures, etc. As far as the promotion component, just putting a strong piece on the board is most of it; how strong is not that crucial. So, if the promotion fraction of a pawn's value is, say, 20%, and the NB is 20% weaker than Q, my pawn is then .2 x .2 = .04 weaker than yours; no big deal. I guess the exception would be if we each promoted, and you'd wind up with the stronger piece. Hmm...

I'm not sure if it is still good, though. You can get into a queen endgame just as well by preserving your queens till the endgame, and I think it would make more sense if you had the same imbalance in both cases. Taking the Clobberers or the Nutters then poses a little risk. Double sharp (talk) 03:16, 25 March 2019 (UTC)

I do wonder about the value of cross-bonuses. If you have a single bishop, and a ferz on the opposite colour, is there a significant bonus? What if the ferz is instead a commoner, which contains the ferz move, but isn't colourbound? I would guess that the presence of other pieces when the board is not jam-packed always benefits the bishop because it allows for some influence on the opposite colour, so this really just confirms Kaufman's comment that we should really penalise every other combination instead of just rewarding the bishop pair (because every other piece pair has some redundancy, but there is none for the bishop pair). I guess the textbook cases of good bishop vs. bad bishop in endgames already suffices as a test for having the minimum influence on the other colour. ^_^ You do always have nonzero influence on the opposite colour because of your king; with Betza's Colourboundmost King (a royal kirin) the problem would be even worse.

My trials with more than 2 Bishops are indeed compatible with the idea that the base value of a Bishop on 8x8 is 350, and that there is a 25cP penalty for not being able to distribute the power equally over both shades. (At least for 0:1 or 2:1 distributions; I did not more extreme cases like 3:1, 3:0 or 4:0; these could involve higher penalties.) When my PC is not tied up in Tenjiku, I will try the following: I will use Pair-o-Max with a 'base variant' on 8x8 that features Q, R3, BW and N. I will then replace one or both BW of one player by BD or FAD, to measure the bonus and cross-bonus for these Rook-class pieces. H.G.Muller (talk) 10:41, 25 March 2019 (UTC)

So it may well be accurate to say that a B is ideally worth more than an N in practice, but that it suffers a weak colourbound penalty unless you have a bishop pair! I'm really looking forward to those results on bonuses and cross-bonuses!

BTW, if you have done some other investigations about colourbound pieces in the past (or will have time for it in the future), I do have a few more questions about colourbound pieces. I presume the bonus of 50 cP is specific to the bishop pair, and that other pairs of colourbound pieces gain more or less. How does the bonus vary as the strength of the piece changes? (Off the top of my head I'd guess it to be something like linear in the value of the piece, so a ferz pair should gain less than 50 cP and a BD or FAD pair should gain more.) Are there also bonuses for pairs of pieces that are not "colourbound" but yet are restricted to half the board, like fbND or fbNA? (Each can only see half the ranks.) And how do bonuses work for even more heavily colourbound pieces? I imagine you must already get some bonus having two DA's on different "meta-colours", as there is no redundancy between them, although having all four must surely be better still. Double sharp (talk) 14:18, 25 March 2019 (UTC)

There is so much that can be investigated, and it all takes so many CPU hours... Come to think of it, when I added the Clobberers to Fairy-Max the piece values might have been partly set by guessing, like that BD is similar to BW (which I had measured), just slightly lower because of color binding, which would be not so noticeable in a pair (and not exactly implementable in Fairy-Max anyway for the lack of pair bonuses in its eval). Fairy-Max is not intended to be top-notch analysis engine, but just a beatable opponent that is fun to play against, so I considered such short-cuts entirely acceptable. Anyway, I have a preliminary result now on BD which gives some food for thought. I hardly get any difference when I play a pair of them on the same or on a different shade, against a pair of BW. (I have to handicap the BW pair by a Pawn to make it approximately even.) This was unexpected. The reason seems to be that the first BD is almost always quickly traded for a BW. The point is that for any intrinsic advantage one has to deduct the cost of maintaining that advantage. To protect a pair you have to adopt a trade-avoiding strategy for that piece type. The stronger the piece, the more this hurts, in two ways: the opponent piece from which you have to shelter it presumably gets stronger too, so you lose a larger fraction of a larger piece value. Sheltering a Bishop from two Knights is one thing, sheltering a BD from two BW quite another. (And it also doesn't help that they are more similar.) OTOH it is not obvious how unequal distribution of power over the shades hurts as the distribution gets more unequal. It might very well be limited by how easy it is for the opponent to withdraw his entire operation on the safer shade. Note that the difference between Bishops on different and the same shade in my measurements has always been the normal pair bonus, although it is much more imbalanced than for N+B.H.G.Muller (talk) 11:10, 26 March 2019 (UTC)

I think this would be a case of the levelling effect again, right? When you have two types of pieces on the board, with values that objectively would be less than half a pawn from each other (e.g. N and B), their values get dragged towards each other, as trade avoidance means that they have to run scared of an opposing piece and not be as effective as they otherwise might be. (I'm not sure how much less than 50 cP the difference must be, but I recall your piece value investigations have shown that in Capablanca chess the N–B and RN–Q difference are both about half a pawn; since their values don't seem to have converged, the threshold must be less than 50 cP. This raises an interesting question about how much the N and the B would be worth if they didn't need to compete against each other, i.e. in setups where the minors are 4 knights vs. 4 knights or 4 bishops vs. 4 bishops.) I agree that cases with different powers on various shades are more difficult; just thinking about it, a pair of B and BD may act somewhat similar to a homogeneous B pair or BD pair, since the weaker piece still has significant influence on its colour, but a pair of B and F would probably be another story altogether as the F is so slow and easy to avoid for the opponent, so I think your conjecture is probably right, even if we don't have quantitative results yet. BTW, what do you mean by N+B in the last sentence? Is it that the difference in total value between 2 B's (same colours) and 2 B's (opposite colours) is greater than the difference between N+B and the normal B pair (opposite colours?) That seems more or less expected since 2 B's on the same colour have greater redundancy than a knight and a bishop. Double sharp (talk) 14:57, 26 March 2019 (UTC)

Indeed, it is all leveling effect. Pair bonuses should be especially sensitive to that, as these are equivalen to having a different value for the first and second piece of a type. But unlike when you have two really different pieces, a piece slightly less valuable than the most-valuable one can now harrass both members of the pair, rather than just the intrinsically strongest piece. (With N+B I meant an alternative to 2B (unlike shade), where there is equal distribution of power over shades, and 2B (same shade), where the distribution is 2:0. With B+N it is 1.5:0.5, (if you count half a Knight on each shade), and it is not clear why 2:0 should not be worse than that.)

Well, I was thinking that there should be some sort of malus for having 0 on one colour if it is 2:0 against 1:1, because then your opponent is a whole piece up on that colour because you have one less piece to oppose him with. This should still hold in the endgame when there aren't any other pieces, as although you can withdraw operations to the colour you have the added strength on, he can just withdraw to the other colour; adding extra bishops on the same colour to a late opposite-colour bishop endgame should not change much, but adding extra knights makes it much less drawish as you can now attack the opponent's weak colour. I don't think K+2B (anti-pair)+pawns vs. K+B (opposite colours)+pawns should be any less drawish than a normal opposite-colours bishop endgame, as the lone bishop can still hold back the pawns, and the extra bishop can do nothing to stop it. Double sharp (talk) 05:09, 31 March 2019 (UTC)

The levelling effect doesn't exclude effective value differences of 50 or 25cP, as we see between the super-pieces. These can be what remains after accounting the leveling effect, when the cost of the trade avoiding is deducted from the intrinsic difference. Only when the cost of absolute trade avoidence is larger than the difference the effective values can get equal, by adoption of a strategy of 'partial trade avoidence' (i.e. accept trading only if there is a sufficiently profitable positional side effect). The question, however, is whether (say) the difference between Q and RN would come out much different when measured against each other (trade-ignoring RN) as when both are measured against BN (trade-avoiding RN). I don't really believe that, or I should have noticed it before. So it seems the leveling effect of the super-pieces on each other is rather small. This is at odds with the idea that a powerful opponent to be avoided would eat a large fraction out of a large piece value. I determine almost exclusively opening values, though. And on a crowded board most squares are controlled already by lighter pieces, and thus interdicted to the super-pieces anyway. An extra opponent super-piece that you have to avoid causes only little additional hardship. That the BW appear to suffer a lot from the opponent BD might have been exacerbated by the fact that I tested them in a context where one of the light pieces (R3) is a 'slow developer', so that in the early middle game there will be relatively many squares that are not controlled by Pawns or light pieces, so that the BW will be the least-valuable attacker. I picked this piece mix to more or less keep the balance between diagonal and orthogonal moves while testing diagonal pieces, and to get rid of the Bishops as a second color-bound pair that might complicate matters. Perhaps I should have gone for WD instead of R3; that would also be a fast developer, reducing the need for the BW or BD to come out (and even actively prevent that through interdiction by the WD), and thus reducing their early trading.

I have completed some more testing now of the BD pair and 'anti-pair' (i.e. on the same shade). The idea was to test them against a pair of pieces that intrinsically were weaker, so that you would even want to avoid trading in the 1:1 situation. Then the first and second member of the pair should suffer the same from the trade avoidance, and their difference (the pair bonus) would not be affected. BW proved significantly stronger than BD, though. (Much more so than I had expected). So I had to weaken it, first to BvWsmW (still too strong), then to BvcWsmW (balanced the anti-pair exactly, but this could be leveling effect) and finally to BmW (which now also lost its mating potential). The pair of BD beat 2 BmW by about the Pawn-odds score (~68%), so I also did games where the BD side received an extra Pawn handicap. This handicap reduced the score by 17.7 to 20.2% (which is within the statistical error for the 300-800 games I had been playing for each material combination). If I then assumed the Pawn-handicapped results should have been exactly 18% higher without the handicap, and averaged it with the result of the basic imbalance, I got advantages of +4.3% for a single BD-BmW imbalance (BD+BmW vs 2BmW), +6.5% for an anti-pair and +17.0% for a pair. Twice the single advantage would be 8.6%, so the pair over-performed by 8.4%, while the ani-pair under-performed by 2.1%. The latter is close to the statistical error bar, but there is no reason why an anti-pair should have no penalty compared to twice a lone piece. In any case it doesn't seem the BD pair bonus is much larger than that for normal Bishops (which is half a Pawn, ~9%).

That's interesting! How much stronger is BW over BD? I wonder how this half-a-pawn value would change for even stronger or weaker pieces; I cannot believe that the bonus for a pair of ferzes would be that high. I wonder how the superpiece values would change later in the game. Naïvely summing their components suggests that the Q and RN gain more than the BN in the endgame because of their rook component. Betza seemed to think at least that the RN was equal to the Q, and that any endgame advantage the Q had was compensated by the RN's incredible perpetual checking activity. Perpetual check is one of the motifs of Q endgames that the R cannot do so well; how good are the BN and RN at it? Double sharp (talk) 05:09, 31 March 2019 (UTC)

In the mean time I have used my laptop to measure R3 in a normal FIDE context. The results are not entirely consistent, though. If I replace the Knights by R3 the R3 have an advantage of 6.5%, about 1/3 of a Pawn, which suggests a value of 340 per R3 (when N=325). There is a caveat that Rook-like pieces seem to be underestimated by about 25cP when starting behind a closed Pawn wall, which could be seen as a positional open-file bonus. When I replace the Bishops by R3 the Bishops win by about 6%, which would be expected because of the pair bonus, except that this comes out a little high (more than 12%, where ~9% is expected). This could have been unfortunate statistics. The worst anomaly occurs when I replace 1 Knight and 1 Bishop; this is expected to produce the same result as replacing 2 Bishops, as it also breaks the pair. But in the first test the R3 even won by about 55%, and when I redid it with more game the result was about 50%. So it seems that the introduction of the R3 somehow makes a Knight fighting on the same side worth more compared to the Bishop.

There may also be a development advantage at play here that weakens the R3, though: in all these setups when R3 replaces a normal minor piece, it would seem to be very difficult to develop and castle on that side. Especially starting the short rook on g1 or b1 seems to force you to create significant pawn weaknesses if you don't want to waste a lot of time developing, so the fact that the R3 looks weaker when it replaces a knight rather than a bishop may simply be due to it being harder to develop from the knights' squares, rather than from some special synergy with the knight rather than the bishop. (Which would be a bit odd, since rook plus bishop is supposed to be a bit better than rook plus knight.) Double sharp (talk) 15:55, 31 March 2019 (UTC)

P.S. I would suspect that because of the R3's shortness, R3+B vs R3+N should be even better for the bishop than R+B vs R+N due to its long range. Double sharp (talk) 03:26, 2 April 2019 (UTC)

Your remark on other forms of color-binding set me thinking. For odd-even rank or file binding you can even have pieces with mating potential! A vRsD has substantial mating potential. The bare King can have a fortress draw when it is on the safe edge file, but when it can be cut off from there (which is possible for about half the starting positions) it is a certain win. (As can be tried on the checkmating applet.) Funny is that the upward-compatible Universal Leaper restricted to odd files would not have such mating potential! H.G.Muller (talk) 09:16, 29 March 2019 (UTC)

I'll have a look at Scirocco, which looks highly interesting and complex at first glance. (At least this quick glance makes me wonder why it hasn't gotten some interest from large shogi variant fans, unless some of the following of those games comes also from their ancient and exotic pedigree in addition to their size and complexity!) I'll also respond to you on Tenjiku evaluation in the appropriate section. Double sharp (talk) 15:48, 24 March 2019 (UTC)

P.S. Rereading Betza's old pages, I think the idea behind the Clobberers and the Rookies is that they were on paper materially superior to FIDE, but had positional inferiorities, because the Clobberers suffer from colourboundedness and the Rookies suffer from slow development and over-concentration on orthogonal movement. Naturally, this worked better for the Clobberers, because colourboundedness is permanent, but positional weaknesses can be offset – and with very few exceptions (passed pawns or king safety), they can never overwhelm material considerations that much. (Compensation for a pawn or the exchange is doable, but you can't have a purely positional queen sacrifice unless it is to create an unstoppable passer – in which case you get your queen back – or to mate the enemy king, in which case the game is soon over.) The fault should always lie with one overstrength minor, because all the queens are not too far from each other in value. So I guess we could weaken FAD to FmaF in the Clobberers, weaken R4 to R3 in the Rookies, and I guess weaken fhNrlbK to something in the Nutters. Based on his article, he was trying to make it worth more like a B than an N, and noted that the rlK part of it is what makes it stronger than a knight. Perhaps fhNbK (which would have seven targets) is better? Or if the mix of N and K is still too potent, perhaps fhNrlbW, which would be a colour alternator and lack mating potential? Double sharp (talk) 04:19, 25 March 2019 (UTC)

I finally found Jörg Knappen's proposal to weaken the Nutters (which IIRC I saw mentioned but never actually saw), in which the charging knights (fhNrlbK) are replaced by drunken knights (fhNrlWbF), removing the backward king move just like the drunken elephant of chu shogi, which destroys mating potential. If this is not enough, he also seems to suggest a further weakening of the colonel (fhNfrlRK) to the charging chancellor (fhNfrlRbK). Given the strong synergy between knight moves and king moves, I am not sure the first replacement alone would be enough; I still feel more comfortable with the weaker fhNbK (so, a ferocious knight, like the ferocious leopard?), but this would need testing. Anyway, experimenting with your checkmating applet tells me that any of the proposed replacements (fhNrlWbF, fhNbK, fhNrlbW) for the charging knight lack mating potential, which may help in the endgame. Double sharp (talk) 07:53, 29 March 2019 (UTC)

Limited-range rook values

I found where you posted them! It looks like R4 is worth about 380 cP, from your last post (7 March 2013) about it; since this is on a scale where R = 485 (i.e. rooks stuck behind a wall of pawns), we should consider it more like 405 cP (so basically 4 pawns, like an overstrength minor). Earlier (27 February 2013) you give the close values, with N = B = 325 cP: WD = 325, R2 = 270, R3 = 345, R4 = 385, R = 485 (all with the 15 cP penalty for starting behind pawns). It seems therefore that R4 was certainly overkill; but with how strong the Rookies are, perhaps they really need not R3, but R2 to balance with FIDE! That cuts off about 2 pawns of advantage, which must be enough to make the Rookies more balanced with FIDE. (Though it's kind of odd to have both WD and R2 in the same army, it is also quite interesting to have all pieces with mating potential despite having an unusually understrength minor!) Double sharp (talk) 21:01, 17 January 2020 (UTC)

This is quite a late reply, but I couldn't help myself: WD and R2 in one army does seem quite strange, and I remember reading somewhere (I think it was on CVP?) that much of the Rookies' opening strength comes from being able to harass the opposing team with the HFD. Would changing it to a HF (too weak) making the H- or D- moves lame or moving it to the rook squares (swapping with the R2/R3) help bring down its strength adequately? Its development is just a bit too easy (although the short rook's development is frustrating to make up for that, and adding an initial F or lame-A move seems too messy), but making the HD moves lame just makes it feel awkward in certain places, and the concept of "exactly 2 or 3 squares but can be blocked on the wazir square" is hard to adjust to. Moving it to the corner just seems right, though; despite the shortrook being more "rooklike" than the HFD, putting SR on c1 and f1 and HFD in the corners corresponds to the power distribution of most some armies. (I realized after typing this that a lot of armies don't actually follow that rule: BD≈FAD in Clobberers, Sausage>Pepperoni in Pizza Kings, oFF>oWoA in Cinders, and the Nutty Knights' furlrurlbakking seems to have gotten its position from being rooklike, rather than being rook-valued. Oops. I still think the HFD should go in the corner, though.) --ChromaTalK 02:29, 25 November 2022 (UTC)

Betza 2.0

Hello, H.G.Muller. I have a question about the Betza's funny notation article you created; In the first revision you wrote there are some descriptions of Betza 2.0 notation, but it shows some notations that the official source on XBoard website does not describe at all, such as 'x' modifier and limiter. Could you tell me how you knew the untold notations? Also, are other sources for Betza 2.0 available on the Internet? --Stepney-I.C.I. (talk) 12:16, 20 December 2020 (UTC)

So far XBoard does not support Betza 2.0; it never got beyond the XBetza stage, and by now its implementation of XBetza is so powerful that there is little incentive to switch it to Betza 2.0. The Interactive Diagrams on chessvariants.com also require XBetza. So I guess it should be considered a mistake that Betza 2.0 is mentioned at all on the XBoard website. It never got beyond the stage of a proposal, so far never implemented anywhere, to my knowledge. The main virtue of Betza 2.0 is that it is more explicit, and thus easier to understand in human reading. For the purpose XBoard uses it this is not relevant, though. H.G.Muller (talk) 15:52, 21 December 2020 (UTC)

Thank you for replying. Do you mean Betza 2.0 has almost the same functionalities as XBetza does in terms of describing how pieces move or capture? Well, I'm little confused by your article about XBetza on chessvariants.com including notations which are not written in the XBoard website. Where are they from? Also, the XBoard website mentions "Extended Betza Notation". It appears to have functions from both Betza 2.0 and Bex notation by David Howe. Do you know anything about this mixed notation? --Stepney-I.C.I. (talk) 18:45, 22 December 2020 (UTC)