GOTM9 - The Japan Campaign site is now active!


Gil Favor's Sidekick
Mar 19, 2002
Colorado, USA
This is my first attempt at a detailed campaign description and interactive reply for Civ3 so I hope you will find it enjoyable and useful. :egypt:

I decided not to just retell the whole story of my GOTM9 experience but wanted instead to focus on the step by step details of one part of the game that I though was very important to my outcome. I also wanted to focus on some very specific parts of the game that I thought would make fairly good strategy and tactics examples for other people to look at.

I know that many people look at the replays and that we also have a number of very nice spoiler posts and websites each month but my hope here was to focus in on the nuts and bolts of how I accomplished something in particular during the game.

I also wanted to try and include some detailed strategy maps as well as some actual combat animation sequences so that's what you'll see on the website. Thunderfall was nice enough to give us space to upload over forty image files as well as twenty pages of documentation in a website type format. We have also linked in six(6) time specific save files so that you can follow along and/or play along with some of the steps in the campaign if you would like.

So here it is in all its glory:

The Japan Campaign from Cracker's GOTM9

Comments and feedback are welcome and you can feel free to PM me or email me through the CFC user CP functions if you have anything personal or specific that you might need. :beer:
I've posted a news annoucement about the site. :goodjob: It's really impressive work! I definitely will take a good look at it. :)
It's...big! :eek: But a job well done too. ;) If someone wants to learn this game, he should read this.

Outstanding job - a really detailed account of the kind of planning and decisions that need to be made during a military campaign.

Hard to imagine that anyone would not improve their strategy and tactics by reading through this.
Excellent work! I really liked how cracker was able to use every little scrap of info in the beginning to make educated guesses about the unexplored terrain and settlement patterns on the Japanese continent.

Indeed, it is so detailed that one can imagine it being a RL war plan...when I read about the 2nd objective of "producing two Great Leaders" for building an army and the Forbidden Palace I laughed out loud, for the spell was broken and I was returned to the civ3 world! :)

One question, though: I noticed in combat analyses that regardless of terrain and fortification status defenders were given an additional 10% combat bonus...where does this come from?
Park Ranger - for some crazy reason they added a 10% bonus on defence, as a blanket amount.

Which makes no sense, because units already have different offensive and defensive values. So there is no need to give a defensive bonus - they could just make the defensive values higher.

It smells somewhat of being a fudge factor they tuned during development to get the combat 'feeling right'. :)
When you look at the combat analysis stuff, notice that I used the cummulative calculation method instead of the additive calculation method for bonuses. Some discussion has indicated that the firaxis programming probably uses the additive method based on prior examples of their games. The cummulative method assumes the defenders have even more of an advantage

It was interesting how close the actual results kept coming to this cummulative interpretation. If I had used the additive formula in the spread sheet I would have really been getting the short end of the RNG stick. :(

On the issue of the 10% base bonus for every defender, I'm not jaded enough to think this is a fudge factor. I just think it is an attempt to say that when two combatants meet in a battle, then the defender always has a slight advantage even when the two units are equal attack-vs-defense.

I am not always sure this logic is historcally correct or accurate but based on direct experience I am not uncomfortable with it. For me its just part of the game, and it is amazing how many players play the game without understanding the combat math and the defense modifiers that are designed into the system.
Well done, well presented, and obviously a lot of work! Yes, it is surprising how often newcomers to the game don't understand the basics of using terrain to their advantage (like not attacking across rivers, not understanding terrain defensive bonuses, etc.) and this page should provide excellent advice. Very nicely done. :goodjob:

OK, now for a few questions. :) I would like to know first of all why you were using primarily regular units and not verteran ones. I'm sure you know that veteran units have a much greater chance to survive and win battles, especially when you factor in the retreat ability. With retreating War Chariots that are only regular, it must have been like fighting with conscripts at times! So... was it just too difficult to get barracks in your cities? I don't think I would have pressed an attack with regular units.

And your map decisions... that seemed like a lot of unnecessary effort to me. I would have traded something to Tokugawa for his territory map, even at the price he asked. It certainly would have made invading a lot easier to know where all the cities were located! Of course I didn't play this game and I'm reluctant to try and second-guess trade decisions when I couldn't see what was going on. But you seemed to have more technology than Tokugawa, so why not send an outdated tech or two for his World Map? But in any case, I'm sure you had your reasons. :)

Nicely played and well done. I know something about the time it takes to put a webpage together myself. :lol:
Originally posted by Sullla
... I would like to know first of all why you were using primarily regular units and not verteran ones.

I know that vet units would have been more successful if I had the luxury of having those type of units. You will see that I had vet and elite warriors and spearmen due to promotions from warrior conflicts, but almost none of these were built as vets.

I was whipping or rushing settlers, temples, spearmen, galleys, and war chariots at every opportunity. Most of them were produced in frontier cities that had no barracks. In despot, I could let a frontier city make one shield, and then whip one person into a war chariot on the next turn.

I think that this "3/3 war chariot thing" is one of the limitations of the Egyptian civ (not that it is a fatal problem). I could have hordes of the 3/3 war chariots and then rely on their movement and retreat ability for their survival.

Also having a large number of 3/3 chariots instead of fewer 4/4 chariots would help with garrison duty and preventing culture flips. Two of the redlined 1/3 chariots were better as garrison than a single 1/4 chariot.

The jungle coverage on the middle islands made it really hard to farm barbs to get the war chariots field promoted. That is one reason I had to sprinkle in some of the horsemen just to have some movement. I was sorta stupid at one point because I whipped out a chariot in a port city surrounded by jungle only to find that I had to send a galley over there to get it out of the town. :crazyeye:
Originally posted by Sullla
And your map decisions... that seemed like a lot of unnecessary effort to me. I would have traded something to Tokugawa for his territory map, even at the price he asked. It certainly would have made invading a lot easier to know where all the cities were located!
Technically I agree with all your points here, maps make life easier.

I was just sorta cranked at Tokugawa from the first second we met. He was angry at me even from the start and that still seems like asking for conflict.

I could have manipulated the map out of him, but after the initial contact, plus seeing the info I bought when I got the embassy I thought I knew alot about the possible terrain. I was also factoring the archipeligo type map plus knowing that Tokugawa was alone on his piece of dirt. If he had shown contact with some other civ I would have probably traded for a map. Another piece of information that I could use was knowing the map was randomly generated by the fractal terrain generator and that would place most of the jungle near the "equator". Since I could see bucket loads of jungle on the central isles (I haaaaaaaaaate jungle) and Tokugawa was near the south pole, I was expecting forests, mountains, and tundra.

The map analysis was more to emphasize looking at the terrain that you can see in the early game and then using your view of the minimap to make some intelligent decisions.

I just played a succession game last month with three newbies (they said they were exerienced Civ2 players) and we started near the south pole with mountains clearly visible to the southwest. Every one of these guys headed their first exploration to the south and into the mountains and then wandered around in the tundra for 60 turns. A little bit of basic terrain reading skills plus looking at the map position would have said "go north" or "go west" and would have lead to grassland, cows, wheat and a winning position.

In the game, I actually made those map decisions in just a few seconds of thought, then it took a lot longer than that to capture the image, scribble the sketch to try and show my thoughts, and write the discussion text.

Very impressive... both the in-game strategy and tactics, and the report.

1. Re the combat calculations, I noticed the cumulative versus additive approach, and actually came to CFC to question the choice. I'm glad to see you addressed in this thread, but I would suggest that you do so in the AAR as well. I think part of your purpose was getting new players up the learning curve faster, and the use of the cumulative approach might be confusing.

Also, do you really think that cumulative approach is correct? Although this specific game matched the more conservative outcomes, everything I've seen conforms to additive.

2. Re Sullla's comments as to barracks, I would suggest a middle path: build Swordsmen and Spearmen at 3 hp, and WCs at 4 hp.

3. Always expect militaristic AI civs to have some loose Archers hanging about, and when at war moved to the front.

4. The WC Army is a mistake I've made (once!). Why, though, didn't you use the elite+leader Swordsman for the Army? That unit is now and forever more compromised anyway.

Again, great job all around.
Originally posted by Theseus
... 4. The WC Army is a mistake I've made (once!). Why, though, didn't you use the elite+leader Swordsman for the Army? That unit is now and forever more compromised anyway.
I think that the specifics of an ancient age army discussion is key to this whole issue.

Unfortunately, many army discussions of this issue get sidetracked by someone who has 17 armies full of tanks or modern armor and that has nothing to do with what we really are talking about with armies in the early game.

I thought for about 17 nanoseconds about putting a swordsman in the army but I was trying to think or the bigger picture. If I put a swordsman in the army it would permanently limit the army to one movement point and eliminate any chance of using the blitz ability to counteract most of the negatives of an army.

I think a key point in framing how the army was used is recognizing that the time period only had galleys for transport (2 capacity) and it would be many turns before caravels or galleons came around. If this issue weighed out to be an overriding concern then it basically meant I could only put one unit in the army for now. This would be to preserve the ability to get the army off the Japan continent so that it could be used somewhere else before the game would be over.

I put the war chariot into the army because it was the closest fast unit and I was suffering from temporary war euphoria/brain damage that had me thinking horsemen and war chariots were equal.

I did not really notice the war chariot flaw until much later in the game when I had ferried the army across the water and then loaded two knights into the army to upgrade it for disemboweling Xerxes. I then had at least two or three instances where the army could not attack a lone Persian defender that was standing in jungle or on a mountain. These events woke me out of my stupor that this was not just a slightly weakened Knight army but it was permentaly trapped in time as a turbocharged warchariot army. This experience permanently etched the decision process in my mind.

Ancient age armies can either be temporary powerhouses that basically become throw away features of combat on one landmass. or you have to think strategically about the chance of using them with added units from the middle ages.

If you look at the results of the combat with 20/20 hindsight, almost all of the swordsmen died. :) If I had loaded one swordsman into the army it would have been destroyed. If I loaded two swordsmen into the army, it would have been the victim of the archer attack in the approach to Kyoto because armies always stick up as the strongest defender instead or being reserved to fit their classified strategy role. The archer would have knocked my 9 hit point army down to 4 or 3 hit points and that would have made it worthless as well.

I general, I am probably advocating a strategy that says you should almost never fill an ancient age army with units unless your are on a large continent and have the Greeks or Romans as your primary opponent. With either of these two opponents, a swordsman equivalent army is one of the only reliable ways to dislodge Hoplite or Legionary defenders.
Cracker: Very nice job.

One little whine though (I never can help myself with this one): Your comment on war chariot retreat odds at one point at the early stages of the campaign indicate a fundamental and very common misunderstanding of probability.

You say that after several successful retreats in a row, the chances of another successful retreat become slimmer. This simply is not the case (assuming no bug or intentional bias in the coding). We are talking non-dependent events here. The probability of a successful retreat has nothing to do with previous success rate. It's 50-50 as always.

It is true, that *before* a series of attacks, you can assume a low probability for several retreats in a row. But, in probabilistics involving non-dependent events, you *never* look back. After three successful retreats the changes of stretching it to seven retreats in a row are the same as those of getting four in a row in the initial situation.

If I misunderstood what you meant, I apologise. For whining over a petty detail I apologise as well (not to mention the terms I used - never discussed maths in English before).

For some ole reason the 'past probabilistics' misconception always ticks me off. This time it made me go through the registering process of another forum... :)

Anyways, the bottom line is, great work! It gave me a few ideas, being an emperor-level player but a rather peaceful one at that. Kudos.

Heikki the weird math geek
Heikki, thanks for pointing that out... I meant to as well.

carver, I don;t know if you spend much time at 'poly, but I've been in (and started) a lot of threads on ancient Armies. I totally agree, never fill'em. But I've become a big fan of creating 2X Sword-level Armies, and later topping them off with a Musketman, then finally an Infantry. As you say, the Archer would have damaged the heck out of it, but think of the hps saved for other units. Anyway, the whole point is to get that first win, and that's easier to do with 3 attack, not to mention that a Sword* is compromised.
Heikki and Theseus,

You are correct that a true random coin toss is always a 50/50 probability but that is not really what we are talking about here.

RNG functions in gaming are not truly random because over n trials the distribution of results will be uniformly distributed across the range of possible outcomes. As n gets larger, the probability that all the outcomes will be in only one half the range gets infinitely small.

Some of this has to do with characteristics of the RNG being used.

The whole arena of Bayesian statistical analysis and stochastic processes would probably give the average civ3 gamer from the target 15 to 21 year old male market segment a complete brain freeze. Classical statisticians and Bayesian's spend their whole lives debating these issues.

The ultimate issue here is that gaming RNGs are not totally random since they produce a string of numbers that will eventually be guaranteed to be uniform in its distribution. If you have produced lots of numbers lower than the 50% level, then eventually you have to draw more numbers above the 50% level. This is part of the 12 decks in the shoe argument wrt to counting aces.

One thing I was trying to do here was to emphasize that key point that in the long run things will average out. Ultimately you cannot be lucky forever and likewise if you don't get killed your bad luck will eventually change to make the average of the outcomes match with the predicted values.

I see. So it wasn't the classical misconception that lead you to your statement on the retreat probabilities. Doubting so *did* feel odd given the quality of the material on your web page. Instead you propose that it's an imperfection in the RNG that leads to this phenomenon.

Not likely, in my most humble opinion. It would be *amazing* if Firaxis, after dozens of man-years in programming the game would screw it up by a RNG that is *that* bad. Granted, the RNG cannot be statistically ideal. However, the level of imperfection must surely be much lower than one that would affect macro-strategies in the way you propose.

The whole 12 decks thing is irrelevant here, because surely, even if the RNG was implemented as a stack of numbers that would be determined in one go and used up one by one, the stack used in Civ3 would be tall enough not to justify any conclusion based on three or four same-way-flips in a row.

Also, íf such a RNG was indeed used (and because of memory capacity issues alone it surely isn't), what you propose assumes that allocation of nice, balanced distribution of numbers to the stack in the first place would be enforced. This would quite obviously render the RNG very imperfect indeed, and hence would not be done.

Heikki the weird computer algorithm geek
OH, I get it now. You're right, it's just like counting cards in blackjack!!

I never thought through the issue of gaming RNGs being forced to balance... so that explains the streaks people talk about, in both directions.

Thanks for the insight.
I have to apologize for getting distracted in the statistics discussion over the last several posts because I was trying to accomplish about 10 things today and this was just one of them.

One point that I wanted to make about the statistics of the combat outcomes was that you have to look at the statistical balance of the "strokes" to get any sort of an understanding of how this actually plays out.

When you look at the strokes in a combat sequenece you rarely see long sequence of events that does not favor the stronger player in the events. The first round of edo combat events is a fairly good example because all of the defenders had the same strength at 1.25 (horsemen or archers in the forest) against mostly war chariots attacking at 2.00. We would have expected that the war chariots would win 61.5% of the strokes and this was almost exactly what happened ober a sample of just 30 strokes. Since these events where being played one righ tafter the other, they provide a fingerprint of the RNG output in sort of a binary win/loss or 1/0 type format.

The longest unbroken win sequence for the chariots was 5 events whiel the longest unbroken win sequence for the Japanese defenders was 3 events. Just about as I would have expected knowing that the chariots would win just less that 2/3rds of the strokes while the japanese would win 1/3rd. Basically the Egyptians chariots should have won twice as many strokes as the Japanese. If we looked at win streaks, the length of the longest Egyptain win streak should be twice as long as the longest Japanese streak. This was just what occurred.

You can actually see these outcomes in the table at:

This just clearly reinforced to me that counting the strokes and the outcomes could be a strong indicator of the expected future outcomes because the RNG actually seems to be so balanced. The streakiness of the actual comabt win/losses comes form how these individual strokes get added up to make a total combat event.

Clearly the more strokes involved in a combat event, the more likely that event will match the predicted statistical outcome. This should come as no surprise because the hit point jockeys that have already jacked up the standard hit point levels for all unit (some as high as the 10 to 13 hit range for elites). have shown that they basically eliminate much of the randomness in the combat system by implementing this change. If they push the hit points high enough, then basically they almost always guarantee that they get the results they want from the combat with minimal risk of a random negative outcome.

Look at the strokes and I think it gives you a better perception of how the RNG really seems to be working.
Each result of the RNG is independent. There is no adjusting. What happens is the initial seed is put into the algorithm and it creates a result. That result is fed back in, ad infinitum. So if you start with the same seed, the results as calculated by a deterministic machine (your computer!) will always be the same. Nevertheless, for all intents and purposes they are completely random, and are indistinguishable from truly random numbers.

That means that the odds of retreat are exactly the same for every attempt. There is no linkage. There is no saving up or luck turning or any of that. If there were, the RNG would be seriously flawed. It is not like a blackjack deck with a limited number of cards. It is a deck with infinitely many cards.

the table you provided link to is big enough a sample to conclude that there are at least no major oddities in the RNG. That's about it. You just cannot draw any detailed conclusions about it from such a relatively small sample.

However, it seems to indicate a distribution that would be expected for a series of compleletely independent events. As I expect it to be as discussed in my post above.

The 'streakiness' is a bit more complex matter and would require a much bigger sample to draw conclusions on. But, once again, the streaks seem about appropriate for good, basic independent RNG. I'd like to know the mathematical foundation on your assumption that streaks for the attackers here should be twice as long as for the defenders.

As Zachriel pointed out, each result drawn from the RNG is independent (for any practical purpose with a limited sized distribution - even a very large one). There simply is no reason at all to think otherwise.

Heikki the weird trusting-firaxis-not-to-make-any-very-basic-mistakes-here geek
Top Bottom