kuroi neko

I did a few tries against the computer. I would say victory will be tied on the fate of both super aces (Sakai on the Japanese side and the other guy in the Hurri II of Death for Britain). Loose one of them (or their wingman) in turn 1 and you can basically kiss the victory goodbye. Also the (random) targets have a huge impact on difficulty: when you get a 3/7 naval convoy to attack with bettys that can do at best 6 damage, it's not the same as a 1/4 vehicle depot which could even be taken out by (lucky) kates . Also the appearance of a small 1/2 convoy makes a lot of difference, it allows to use the kates with a good chance of success or choose to leave a couple of points to the Brits and pick the best fighters on air superiority, while a game with no weaker objectives than 1/4 will practically force the Japanese to loose or draw a bombing run (tried once to bomb 3 times in a row with the bettys, they died trying ). As far as playing against bots is representative, I would say the Japanese score will be more centered around 10 points, even with a massive failure, while the British may range from 5 to 20 depending on the objectives and air superiority missions (the 20 points were when I was lucky enough to kill Sakai first turn ).

I guess the difference between the demo and full is the memory usage. On my system (XP SP2, athlon 2800+ 1GB ram), the demo game & lobby clients use about 20 Mb each while the retail version the lobby needs about 50Mb and the game about 60Mb (the exact amount varies constantly by about +- 3 Mb). I get very rare connection failures at start. My guess is that the system takes too much time swapping the client back into memory, thus causing a timeout. Since I get a confortable amount of RAM, I assume the problem could be more frequent on 512 or even 256 Mb systems, and also such programs as anti-viruses (especially coupled with mail clients) could cause a lot of memory swap. To minimize the problem, I changed the settings of my mailer to check for mail less often, thus leaving the anti-virus mail check inactive most of the time and reducing the memory strain. Just my 2 cents...

Ah well my bad, sorry . Proper number of cards indeed: 4(attack)+1(teamwork)+2(defense)+1(check six). Medium alt for b17s is pretty rare, and against such killer German planes it is indeed a suicide run. Too bad I never had the fortresses flying that low when I was playing for the other side . BUT I still cant understand: - how the system is supposed to produce such a number of IMS2:D and OTS3:4 in a row (possibly warping probabilities, which may explain the frustrating results - losing a wingman and a bomber to IMS2:D while countering a third one, plus an ace wasted countering an OTS3:4 and another bomber crippled by two OTS3:4, all in 3 turns, is a bit annoying) - why the bots are allowed to draw more than one card in a reaction phase (both in attack and defence, just try to attack a wingman and see him add a card to his hand each time you play a manoeuver or burst against him) - why the number of bot's extra cards often exceeds what is stated on the pre-flight screen (pretty hard to pinpoint this one, hence the idea of a systematic measurement on the server).

Ok, I tried the alied '44 bombing campaign and I must say the usual AI redraw bug made a complete mess out of it. Basically it's a picnic ride on the German side, and a tedious, frustrating ordeal on the Allied side. The main reason: the obvious AI redraw bug, which turns the AI fighters into mindless player and bomber killers. Especially the wingmen who can redraw a preposterous number of cards. As it is now, the side owning bombers is at a huge disadvantage, and it shows immediately. I managed to wipe the floor with the Allied (something like 110 points to 55), but as Allied I simply cant save a couple of rested B17 escorted by my best Thunderbolt pilots agains a pair of fatigued fw190. A striking example : I found a really brillant fw190 wingie who started a turn with no less than 8 (eight) cards in hand. Here are the characteristics of the German "ace" Name : Norbert Haff Skills : Natural pilot Teamwork Check six Anticipation Distract wingman Sweep tail Draw an extra card x2 Target pilot x2 fatigued (one less structure point) Altitude : medium attacking a B17 As far as I understand, the fw190 wingman should draw 4 attack cards at medium. The teamwork skill should add one for a total of 5. All the other skills have no influence on the number of wingman attack cards, as far as I understood. Assuming the AI drew a card at start of turn we could rise to 6. Now as it appears, the wingie, not satisfied with this meager hand, drew yet another card after playing 2 or 3, for a grand total of 9 (nine) cards in one single turn. And of course he had already drawn another card at turn one, which means he managed to summon 3 cards out of thin air. At any rate, it is quite infuriating to see the bots draw as much cards as they want in a single turn, while the human players (and the bombers it seem) are limited to one extra draw in that situation. Or if they aren't I would like to understand how the interface allows to draw more than one extra card in a reaction phase! I reckon this problem was not so apparent in the demo due to the modest plane performances, but experiencing the difference between drawing for instance 6 and 9 cards in one turn is really an eye-opener. For very rare cards, the probability is roughly a linear function of the number of draws. For instance, the probability to get an IMS2:D from a complete deck of 76 crads in 6 draws is about 7.9%, and it rises to 11.8% with 9 draws. Now the probability to get at least an IMS2:D or OTS3:4 will rise from 15.3% to 22.3% (computations done by the book, using the statistical "hypergeometrical law" as we called it in my French preppie days ). Which means the redraw bug is increasing the probability of appearance of a "sudden death" card by about 50%. Now the actual "sudden death" probability is also increased by the fact that the redraw time is chosen to optimize the efficiency of the killer cards (the bots will select the oponent with the smallest hand). It means you practically stand no chance once the wingie has drawn the killer card if it is accompanied by a random choice of other helper cards that seriously outnumber the cards in your own hand. A player wingie will NEVER get 9 cards in a reaction phase, more like 3 or 4 for comparable planes. The difference in numbers will practically make the killer cards unstoppable. Where there is still a chance of defending against 6 cards with 4 cards, there is a huge lot less against 9. I would say a 9 cards hand against a wingie defending with 4 random cards has about 15% chance of ending in sudden death by IMS2:D or 4 points of damage from an OTS3:4. Not counting the possible damage with conventional IMS cards or weaker OTS. I've not done the exact maths but it should rise the sudden death probabilities to above 20-30%. Deadlier than Russian roulette . Similar considerations apply to the bombers, but I have no idea of the bomber reaction cards system, so... And also, the ability for the bots to draw multiple cards in a single phase as they see fit is very seriously tweaking the probabilities: drawing a new card each time a dangerous card is played is of course a much more optimal use of extra draws than what the player is allowed to do. This bug is a complete fun killer. I may understand the AI planes are biased in dogfight mode to prevent people from gaining easy experience, but in campaign games at least, I would like to see bots that abide by the rules! Ah, and besides I saw a campaign game with no less than 3 IMS2:D and 3 OTS3:4 in 3 turns. The cards drawing code IS buggy, either in the principle or implementation, and after playing around with the full version I have lost any remaining doubts about it. I suggest again to add a statistical measurement system on the server, to get the real repartition of cards and count the actual number of redraws. This should be a great debugging helper and also a non-regression tool for future evolutions.

First, thanks for taking a lot of our suggestions into account. The keyboard shortcuts are quite handy, as is the leader cards window. However there is always room for improvement, so here are a few suggestions : - it would be nice to have a visual clue for the private/broadcast mode of the in-game chat, like a red lamp lit when private mode is on. Also, the ***SECRET*** indication is a bit conspicuous, maybe a color change similar to the server messages in the lobby chat would be more convenient. - for some reason the game window grew bigger (at least in my 1920x1440 resolution). It would be nice to have the window size configurable as an option, I rather preffered the smaller one since it left more room for the lobby chat window - the leader cards window has to be opened at the begining of each game or when the wingman takes over, it would be more convenient to memorize its position and have it opened automatically (possibly as an configuration option too). - trying to use the "options" menu of the game window currently causes an error. I assume this is a just demo problem? - the "away" switch still does not work 100%. Small annoyance but still... - it would be nice to have a fix & changes summary with the new releases, if only to know about remaining issues in order not to report them again and again. But all in all this is a good release, I saw no stability problems and the game is still great .

At last Egao will have to occasion to die in the cockipt of a Zero . Thanks for the update!

That'll be an extra stripe for you, squaddie .

Thanks a lot. I hope my remarks didnt seem too harsh, I made them out of love for the game .

Ok, just an observation about cards drawing code: I had an IMS2:D in my hand at turn one. Played it, got another at the first redraw. This is clearly not a reshuffle issue since the deck of 76 cards could not have been emptied at the time. Now if the logic allows to redraw any discarded card, then this is statistically catastrophic : the most interresting cards are usually played quicker than weak ones. Especially the attacks, played immediately by wingmen. So putting these cards back into the deck immediately will increase the chance of getting them back sooner than the usual deck system would allow. The net bias toward "hot attacks cards" will increase tremendously, considering the very limited probability of occurence of the IMS2:D, OTS, etc when usual drawing order is used. I guess this also could explain the preposterous "luck" of bots with tons of redraws : as soon as a good serie of cards gets played, it has a good chance of being "recycled" immediately, giving an exponential advantage to the pilots who can redraw. And by the way, I noticed countless times bots drawing more extra cards than listed on the briefing screen. More precisely, when I counter a move from a wingman, he gets to draw a new card each time I play one. So far so good, but when I count the actual number of redraws, it looks more like the redraw limit is applied only once per turn. In other words, if the AI has 4 redraws at turn one, it will draw up to 4 cards at turn 1, up to 3 at turn 2, and so on, as if the max redraw count was only decremented once at the end of the turn instead of after each redraw. Is that a special "player-killer" feature for demo limitation, me not understanding what "draw an extra card" does, or a general problem with redraw limitation logic?

Yes sure, this was just a suggestion. No hurry at all .

Wile I am at it, would it be possible to have a "highlight" system, like on most IRC clients? The idea is to have a list of keywords that would be displayed in a distinctive color (red or something) and trigger a (configurable) sound notification each time they appear in the chat (in the messages, not the player's names). Usually the main keyword will be the player's name, but it can be anything really. This would allow people to notice others requesting their attention.

Problem solved, and in no time. That's one more cross behind the cockpit for you. Thanks a lot .

otetsuki tokubetsu kôkûtai has currently only 9 active elements, but I cant add more. Looks like the server thinks I have 12 active elements. 3 leaders were indeed killed, looks like they were not counted out or something. Could you have a look at that, I would be sorry to let people wait at the door .

Shameless bump... Games with 3 OTS3:4 or IMS2:D seem to be a common occurence, with Pawlock and Sixxkiller we witnessed about half a dozen now. Now the games with something like 2 IMS2:D and 2 OTS3:4 in quick succession are much, much more frequent. I would say maybe one game out of 3 or 4, but of course this is subjective, measurement would have to be done. Really something has to be done about it. In the hands of a leader, an IMS2:D+Ace pilot combo is a guaranteed fun killer. Well, wingie killer to be precise, but in turn 1 or 2 the fun usually follows the wingie down the drink. In the hands of a wingman, more than a single OTS3:4 or IMS2:D per game may soon become enraging for the victim and not so satisfying for the winner (if he values smart decisions over blind luck). Frankly I dont think the increased number of cards (extra or redraws) in the full release will be enough to correct this apparent bias toward more randomness in comparison with the original card game. As far as I understand it, IMS2:D or OTS3:4 are meant to keep you on your toes, with the downside of having this game-killing capability. Now if they can appear like 2 or 3 times in a game as wingie's free potshots, able to smoke or kill a plane with the only tactical merit to have bought "team work" to maximize the chances of getting these presents from lady luck, I think it's a strong deviation from the original game concept and it should be corrected to revert to the original intend. Maybe implementing the exact simulation of a deck of cards will be the solution, at least the closest from the original, proven design. So much the better for card-counting players, but at least casual players would curse the computer less often On the technical side of things, it appears cards drawing does not work as expected and/or explained. First, I suspect a bug in the drawing routine. I would suggest a test : on the server, record all the cards drawn in every game (by batches of 76 cards) and verify that : - exactly the 76 cards of the deck appear in a serie - the statistical repartition of cards matches the expected percentages Second, using the standard rand() random generator is dangerous, since it is well known for its limited statistical reliability, especially when used in very repetitive patterns like cards drawing. Depending on how the random seed is initialized, the system may produce very few distributions of cards among the possible factorial(76) ones.

here is a rather compact wikipedia article on the subject. Basically it is a point-based system that takes the "expected performance" of a player into account. If you get a rather high score, you will be expected to win more and more against weaker players, and the penalty for loosing or even getting a draw against weaker players will increase accordingly. This will tend to stabilize your score around your "real" skill level, since early victories wont prevent you from loosing points for mediocre performances against weaker players. Conversely, a single victory against a strongly rated player will be very rewarding since the inital score difference will be interpreted as an expectation for you to loose the game, yelding increased reward in case of victory. To get an initial ranking, the score is allowed to evolve quickly for the first few games (5 or so). Then a "dampening" effect kicks in and the average score tends to stabilize more and more over the next games. In a nutshell, this is what allows to reduce the advantage of playing many games in terms of global score : many mediocre games will drag you down the ladder, while a few brillant ones will push you up far more quickly. A few years ago I was playing the Close Combat III ELO ranking system, and I was rather impressed by the accuracy of the results. After a hundred games or so, I was hovering around a given rank and I reckoned by and large it was a pretty accurate reflection of my real skill (i.e. better ranked players were indeed better than me, and lower ones usually less successfull). In the end I even took advantage of this ranking system to challenge other players who had about the same score, it was a way to get interresting and balanced games.

I would say keep track of two different scores : 1) the best single mission XP total 2) the best average, but pruning people with less than 5 missions or so (so that the average reflects constant performance rather than a couple of lucky games). The same could be done for kills/mission, although single best mission would be awarded to the first lucky guy drawing enough ammo to down 3 or 4 planes in a single game . I guess the "5 missions or more" filter would make more sense there. There is also a strong bias, at least in the demo, due to different planes characteristics. Obviously p40s usually get better kill ratios than Me109s. Speaking of which, I also would like to see a "kill ratio" ranking (number of kills/number of bail outs), i.e. taking survival into account. After all, it is more difficult to get high kill total if you value survival in the long run. That could be an incentive for less reckless flying . The "ideal" system would be something like the chess ELO ranking. I guess the maths are not too complicated to implement on the server, and as far as I know this is the system that best compensate for uneven number of games played. [ September 23, 2005, 01:29 PM: Message edited by: kuroi neko ]

For people like me who stay connected for long periods, it would be nice to have the local time added in front of the chat messages, like for instance: <12:23 kuroi neko>anyone for a 2v2? <14:28 john doe>oops too late

luck of the draw

Any idea of the shipping delay? Like for instance the extraneous time to reach the Land Of The Frog Eaters? .

I had connection trouble a few time, and I noticed it was always while my PC was doing a lot of disk swap (I have a few other applications in te background, and often a handful of browser windows). I wonder if others noticed a lot of disk activity during these connection problems?

Mmmm... I don't know the total number of OTS3:4 in the deck but since it's rarity is rated "1%" I assume it could not be more than 2, and I actually saw 2 in my own hands AND fell victim to a third in the same game (only one at a time, as I was in a situation where I could do little but discard them). At the time I simply assumed there were 2 in a deck, but now I wonder... Is it just me, or do you also smell a slightly fishy scent emanating from the general direction of the random picking code? . At any rate, it gives a completely new meaning to our "wrong cards picking special air force" .

I spend rather a lot of time (far too much, to say the truth ) connected to the lobby and got to know a few individuals. Therefore I especially try to let them know if I'm actually available or not, although it's already basic good manners even to unknown players. This is how I noticed the "away" function was a bit awkward to use. It appears the "away" menu toggle does not follow the actual "away" state in several occasions: 1) when reconnecting to the lobby after a provider disconnect, the displayed status is "present" regardless of the menu settings. If you were away before disconnection, you will have to toggle the status twice to return to a proper away state. 2) if you play a game, the away status is switched off but here again the menu option is still set. Beside the minor inconvenience of having to switch it back on, I wonder why playing a game for instance against bots or someone you happened to chat with by other means should lift the global "away" status. This is of course a minor glitch, but I don't like to be rude to other players by appearing active in the lobby while actually AFK. For now it's really easy to forget about the away status and leave for hours while still connected and displayed as active, thus bothering other players by seemingly ignoring their greetings or questions.

Good point .

Mmmm I was thinking of something maybe completely crazy, but for the sake of the argument, I'll go ahead regardless . What about storing all pilot's moves, the whole deck of cards in the actual order they appeared, even pre-computing next draws (this should be possible provided the random number generator state is saved at some point), in short gathering all necessary parameters to replay the exact same game up to the disconnect point? Now let's say a game crashes or disconnects. One could let the players resume it at the exact same point. Problem solved, if and only if each human player agrees: the game resumes as if nothing happened, except the inconvenience of restarting it. I have not thought in deep about the details, but It seems to me designing an automatic resume system should be possible: the server would wait (up to a reasonable delay, like a few minutes) for all disconnected players to reappear in the lobby (in case the disconnection was due to a client loosing internet access completely, otherwise the game disconnect should not even affect the lobby connection), then prompt them for restarting. Once they would be all back online and have clicked the "resume" button, the resumed game would launch automatically. Of course, a player declining to rejoin the resumed game would have to face the consequences in terms of popularity, reputation or whatever. If it's a plain technical problem it will be obvious (the guy will vanish from the lobby or at least wont be able to play for a while), but if he can be seen happily playing other games a few minutes after, well... Now assuming some player cannot or doesnt want to resume the game with the other humans, his element will get stuck with this game to finish (against bots) before being allowed to play on. At this point, several copies of the same game would have to be kept (one for each group of human players playing the end of the game separately, i.e. a maximum of 4 copies for a 2v2 game with all players human, but still only one copy of a given game per player's element, which is enough to guarantee a non-proliferation of game state copies ). If some player resumes and plays but a few more moves before getting disconnected again, they are recorded by the same mechanism, and the resume point simply advances accordingly. Not much left to do than play the game to its very end, however bitter it may be . Of course, letting bots finish a game started against a skilled human may improve your chances of survival, but this choice would have to be made in broad daylight . Maybe just whistful thinking, but still... [ September 13, 2005, 06:32 PM: Message edited by: kuroi neko ]

Bah sc<censoded>w realism, the fact that a wingman could be let loose with a fresh hand of 6 cards regardless of altitude is just a very fun and interresting tactical parameter to be taken into account. On the defensive, it makes climbing worth the limited cards and redraws, and on the offensive it forces to do more than simply pour lead into the leader hoping to get better draws. After quite a few games against humans, I noticed the "instant wingman revenge" tended to occur less and less often. Most of the time the player with the advantage would adopt more refined strategies, like letting his wingman get the kill on the leader to be able to climb out of reach with his own leader, damage the wingman before going after the leader, etc. IMHO, the "fresh wingman threat" is a major element of the game strategy. It offsets the huge loss of the leader with a possibility for the loosing pilot to spend the rest of the game doing more than escaping a pair of bloodthirsty maniacs (although this may also happen from time to time ). The presence of this threat will force the advantaged player to implement careful risk management, unless he wants his leader to experience a really short-lived triumph . I would rather let it stay as it is.