Restart eliminated players: How it works (and how to fight it)

[SlowThinker]

'Existing cities' - do you mean all existing cities or cities of color that is respawned?
'equaly remote' - do you mean something like gravity center?

[Peaster]

ST - I think he means all existing cities. Not including the respawn color, which has no cities.

My theory about the algorithm is that it looks randomly for a location that is at least N = 25 squares away from all cities. If that fails, it tries again with N = 24, etc, until it succeeds (or decides randomly to quit). This would produce the "equally remote" pattern solo observed.

[solo]

Yes, as Peaster says. The location of the civ just eliminated does not factor in except for the fact that its conquered cities are part of the group of all existing cities.

[SlowThinker]

Do I understand correctly? The theory is:

The Civ2 program remembers position of all cities of all colors of the beginning of the turn when a civ was destroyed.
Then it takes all squares that have distance from nearest city >=25. A respawn position is chosen randomly from these candidates.
If there is no square with distance from nearest city >=25 then squares with distance from nearest city =24 are considered in next step. Again A respawn position is chosen randomly from these candidates.
If no such a candidate exists then 23 is taken and so on.

[solo]

The determination is made after the civ is destroyed. If its last city is razed, this could affect the resulting respawn location.

The rest of the theory is just a guess based on observations. From my repetitive tests, done by reloading the game over and over, please note that the final respawn location is usually not on the same tile, although many do appear in the same areas. This suggests a slight random factor is involved, too.

Best thing to do in a game is build MPE and give tech, etc. to respawned civs so maps can be traded to pinpoint their locations.

[SlowThinker]

Thx for an explanation.
Does annihilation of all civs in one turn lower chances of respawn?

[solo]

I doubt it, since killing two in one turn in my tests did not reduce respawn chances.

[SlowThinker]

How close may respawns be to existing cities? I had respawns at distance 13 and 14 in Holidays Tournament and I have a suspicion that there is some limit, maybe 10, 12?
In other words you get no respawn if you fill the map by cities densely enough(?)

[solo]

Originally posted by Zenon

Some remarks on other theories that "respawn" from time to time: killing all civs in the same turn does not prevent respawning. All tests indicate it. Respawned civs can appear right next to your troops and there is no indication that your units influence respawns. Respawns are likely to appear far from other cities, but I have seen them appear within city limits when the map was occupied by many cities.

Zenon

Here is one answer to your questions. My own experience confirms that respawns often are next to my units that have just nailed a previous respawn. (no map changes in time to change the respawn area)

[Andu Indorin]

All of these observations are most interesting, enough that I've begun a quick study of the problem. Operating on the assumption that restarts are influenced by the availability of suitable restart site, my initial study used a "restart" friendly custom map.

While the study is no where complete, based on 14 tests utilizing 6 civilzations each, I can say that one number repeatedly occurs: the base chance for a civ to restart is appears to be a function of 47%. On the first restart, the chance (2 x 47%); thereafter, subsequent restarts occur 47% of the time. Many of you will no doubt recognize that this percentage corresponds to the "Continent Rule" from Samson's Study of Hut Outcomes. There is certain coherence between the chance of a civ restart and a hut outcome; after all, in sense, a restarted civilization is much the same as nomad/advanced tribe obtained as a hut outcome.

If this basic percentage holds with continued testing, we might view it as the first pass/fail condition of a civ restart. If an eliminated civ passes this first restart condition, then a second test must be passed: achieving a suitable restart site. The challenge, of course, remains in figuring out how this second condition is determined.

[SlowThinker]

Suitable restart site ...? Do you mean a square or an area? I got a restart on a Mountain square...

[Andu Indorin]

In terms of suitable restart sites, I was initially thinking along the same programming logic revealed by Samson's "near city rule" in terms of hut outcomes; in the theory, if the restarts of AI Civs in Civ2 represents a programming parallel to the logic of hut outcomes ... after succeeding an initial "do-or-die" restart check, an AI civ would still have to pass a second check to see if there is still space left in the world to pass -- at the very least -- the near city rule. I, too, have seen AI restarts wind in the strangest places. It's been a while, but one of the more interesting outcomes I've seen along these lines are three or four settlers from different civs restarting on the same island, with no grassland/plains whatsoever to speak of, and maybe seven or eight squares to work with; if we could determine from this one extreme (reductio ab absurdum) example how Civ2 programming allowed such an outcome, we'd be one step closer to the wiser. (ps. in my game, if I recall correctly, I had two AI settlers stuck there ...)

But, if we can determine a statistical base probability for a Civ restart, I'm thinking that the next question is: what determines WHERE a civ will TRY to reestablish itself ... i.e., what is criteria for failing the second half of the "Re-Start" equation.

[Old n Slow]

Do restarts ever occur on the poles? What is the smallest sized isle? In a recent GOTM, the world was largely jungle, and yes restarts would appear on small isles with nothing but jungle to work with.

Perhaps that map (with the many annotations) might be a resonable test environment to play with.

[Andu Indorin]

Summary of Initial Findings on AI Restarts

(with due apologies for the delay -- had to resolve the "split AI civ restart" phenomenon ... ).

All tests performed with 2.42 Classic, at Deity and on a premade, all-land Normal Size map (to ensure that terrain considerations, as a possible variable, does not effect the effort to determine base percentage chances for restarts.

"Exploratory Study" of restarts done start at 4000 bc (turn 0), 1550 bc (turn 50), and 25 bc (turn 100); one start with each color, each start with 6 AI civs, yielding 42 civs for each test "run." Initial results for first two "runs" indicated a c. 95% restart on first two restarts, dropping to c. 47% thereafter. Results for third "run," however, yielded a 33% restart on the 4th and subsequent AI civ restarts. Thus, AI civ restarts do appear dependent upon the game turn.

Operating on the assumption that the restart percentages may potentially change with changes in the "Advanced" restarts (i.e., a greater than size one city restart), the effort is made to track down these thresholds:

Turns 1 to 20: One settler yields Size 1 restart.
Turns 21 to 40: One settler yields Size 1 restart. A second settler becomes available for this and all later restarts.
Turns 41 to 80: One settler yields Size 2 restart. For this and all later starts: city starts with Temple, Market, and Library; AI civ may receive one or more military units; and irrigable terrain (grass, plains, desert, tundra, and hills) adjacent to the new city is irrigated (see below).
Turns 81 to 100: Size 3 restart.
Turns 101 to 120: Size 4 restart.
Turns 121 to 140: Size 5 restart.
Turns 141 to 160: Size 6 restart.
Turns 161 to 175: Size 7 restart.
Turn 175: at deity (1500 ad), no further restarts.

NOTE: Both AI restarts at Size 2 and above, as well as "bonus" irrigation on the restart are dependent upon the Human player's "geographic" knowledge. Bonus irrigation only occurs in those squares that haven't been revealed on the Human player's map (i.e., still "black"). Similarly, if the city site of the restarted AI is in a square already "discovered" by the Human player, the restart is limited to a Size 1 city, with no city improvements. (Coincidently, it was also discovered that building the Apollo Program also prevents AI civs from restarting ...)

Assuming that the percentages chances for a restart will be consistent with the intervals above, same basic test performed in the "exploratory" phase performed for each of the above intervals, beginning at the midpoint of each interval (turns 11, 31, 71, 91, etc.) plus an additional test at turn 51, yielding test starts at 20 intervals. One start per color and 6 AI civs in each game, each test interval begins with 42 starting civs; with nine such intervals, the starting n = 378 AI civs. Interpretation of the results yields the following base chances for an AI restart:

On tests at turns 11, 31, and 51:

Chance of 1st and 2nd restarts: 97.6%
Chance of 3rd and subsequent restarts: 46.9%

On six tests 71 thru 171:
Chance of 1st and 2nd restarts: 95.9%
Chance of 3rd and subsequent restarts: 34.2%

Apparently around turn 61 (1000 bc), the percentage chance for an AI restarts drops on 3rd and subsequent restarts; this may explain why there is no "apparent" intervening, twenty turn interval between turns 41 and 80.

NOTE: It is imperative to remember that in these tests, the effect of advanced technologies are not considered. Preliminary tests using some of the "threshold" technologies (Invention, Industrialization, etc.) suggests no influence on the first restart, but an apparent reduction in the percentage chance of 2nd (and presumably) later restarts. Soooo, the next step will be to determine exactly which techs alter the percentage chances of a restart; and, how these "key" techs cumulatively reduce the chances of restarts in relation to the number of turns played.

Originally posted by Old n Slow
Do restarts ever occur on the poles?

Apparently not. Nor did any AI restarts occur on any squares immediately adjacent to the northern/southern polar icecaps; hence, we might conclude that an AI civ will not restart without a full city grid available. But AI Civs will restart with two northernmost/southernmost squares in polar icecaps.


Finally, on the "Split AI Civ Restart" phenomenon. As mentioned above, it took me a while to figure out the workings of the split AI civ restart phenomenon; i.e., those case when an AI civ will start not with one but two cities, often located at great distances from one another. I've provisionally figured out that this occurs when (a) the AI civ's last city -- at size 1 -- is either destroyed through combat or occupation; and (b) the AI still has a NON settler. For the destruction of the city, the AI civilization is destroyed by the Human player (or, presumably another AI civ); with the lost of the NON settler, the AI civilization is destroyed by the Barbarians. This is credited as two "restart opportunities"; if both of these opportunities yield successful restart results, then two cities will be founded instead of one.

[Elephant]

Your first note really clicked with me. I had a game where the Zulus were wiped out and restarted as the Babs. I contacted them via MPE, traded techs for maps, and discovered they had a size 3 city with irrigation all around - in one turn. They also had a pile of "free/starter" techs (if you are still researching, take a look at those starting techs as well) that made their initial research fairly cheap. I kind of wonder whether they get a set number or an average of the other civs. They had restarted in the "black", but other times I have seen restarts on terrain I've explored they have not done as well. So if you want a decent restarted research/trading partner, don't explore much of the unknown map, but if you want to hobble restarts for faster conquest try to explore as much as you can. Makes Explorers even more valuable for Early Conquest games.