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

[solo]

Respawn Test of One Color

In this test I worked from a save created just before the original purple civ, the Mongols, were destroyed. I replayed this turn 100 times, to get an estimate of the probabilities for respawns. This test was made using Civ 2.42. The results of each test are listed below:

nr = no respawn

When there were respawns, new purple civs were destroyed as long as they kept appearing.

For each respawn, the civ name and respawn location (in map coordinates) are given.

1. nr
2. nr
3. Sioux(18,64) – nr
4. Indians(13,65) – nr
5. nr
6. nr
7. nr
8. nr
9. nr
10. nr
11. nr
12. Sioux(8,70) – Indians(9,73) – nr
13. Sioux(19,65) – nr
14. nr
15. Sioux(18,64) – nr
16. nr
17. nr
18. nr
19. Sioux(13,65) – Indians(13,65) – nr
20. nr
21. Indians(18,64) – nr
22. Indians(12,66) – nr
23. Sioux(13,65) – nr
24. Indians(9,75) – Sioux(13,67) – nr
25. nr
26. nr
27. Indians(13,65) – nr
28. nr
29. nr
30. nr
31. Sioux(8,76) – nr
32. Indians(56,44) – nr
33. nr
34. nr
35. nr
36. nr
37. Sioux(12,66) – nr
38. Sioux(13,67) – nr
39. Sioux(13,65) – nr
40. nr
41. nr
42. nr
43. nr
44. Sioux(13,65) – nr
45. nr
46. nr
47. Sioux(11,65) – nr
48. nr
49. Sioux(12,66) – nr
50. Sioux(12,64) – nr
51. Indians(13,67) – nr
52. Sioux(18,64) – Indians(13,67) – nr
53. nr
54. nr
55. nr
56. nr
57. Indians(13,67) – Sioux(13,67) – nr
58. nr
59. nr
60. nr
61. nr
62. Sioux(13,67) – nr
63. Sioux(13,67) – nr
64. nr
65. nr
66. Indians(19,65) – nr
67. Indians(19,65) – Sioux(13,65) – nr
68. Sioux(13,65) – Indians(18,66) – nr
69. nr
70. nr
71. Sioux(9,69) – Indians(13,67) – nr
72. nr
73. nr
74. nr
75. Sioux(9,75) – nr
76. nr
77. nr
78. nr
79. Indians(13,67) – Sioux(12,64) – nr
80. nr
81. nr
82. nr
83. Indians(57,43) – nr
84. nr
85. Indians(8,76) – Sioux(58,44) – Mongols(12,66) – nr
86. nr
87. nr
88. Indians(10,68) – nr
89. nr
90. nr
91. nr
92. nr
93. nr
94. nr
95. Sioux(59,43) – nr
96. nr
97. nr
98. nr
99. nr
100. Sioux(13,67) – nr


After the test I tabulated the results:


Results after destroying the Mongols:

nr – 64 times (64%)
Indians – 14 times (14%)
Sioux – 22 times (22%)
Mongols again – 0 times (0%)

It looks like the odds of a respawn occurring after killing the original civ are about 1 in 3.


Results after the first respawn settler was killed:

nr – 26 times (72%)
remaining purple civ – 10 times (28%)

Although this test sample is not as large, it appears odds of a second respawn are slightly less than 1 in 3.


Only 1 second respawn was recorded in this test, which was only 1 out of the ten attempts. This second respawn yielded a Mongol, the orginal purple civ. At this point a save was made and another test was done using this new Mongol settler as the first victim.

50 trials were made and the results were:

nr – 46 times (92%)
Indians – 3 times (6%)
Sioux – 1 time (2%)

This result suggests that a single color is quite unlikely to have more than 3 respawns.



I also tabulated the frequency of respawn locations, which were:

(13,67) – 11 times
(13,65) – 9 times
(18,64) – 4 times
(12,66) – 4 times
(19,65) – 3 times
(12,64) – 2 times
(9,75) – 2 times
(8,76) – 2 times
others – just once

All but 4 of these 100 locations were clustered in just one part of the map. Twice, the second respawn was in the identical spot. This confirms our experience while playing. The 4 that were not in this cluster formed another cluster of their own, but at quite a distance from the main group. This suggests that the algorithm for locating a respawn is consistent, but that it includes enough randomization to make sure the exact same spot is not chosen every time. Included is a very slight chance that the area chosen may not be the one detected as the most desirable.

Below is an image I made of purple settlers in the locations chosen by the algorithm for 96 of the 100 settlers. The Greek trireme points at hex (13,65), which was chosen 9 times for the respawn settler. Right below it is (13,67), the favorite respawn spot.

I plan to run another test to see what happens when more than one civ is destroyed on one turn. Results of that will come later.

[debeest]

Science marches on! Thanks, solo, for taking the time to research this.

One of the things I love about Civ is that the algorithms are so complex, mimicking real life. You can go through life and never see the patterns, or you can pay attention and notice somethings (when I chew on this willow bark, things don't seem to hurt so badly), or you can isolate components of a situation and replicate them and identify the underlying component causes and learn to control them.

It's beautiful.

[Bloody Monk]

Excellent work, Solo!!

I wonder if the results are affected by the test subject being the Purple Civ?? If as you suggest, game algorithms are involved, then these might be set from the beginning and, therefore, create the same problem of not even getting an original Purple Civ. In other words, could there be different possibilities arising from different Civ colors??

Not to complicate things....

Monk

edit: I meant to ask, what was the original coordinants of the Mongols?? Thx

[solo]

Monk,

Second test follows with two new colors, with conquered cities in much different area than the purple civ of the first test.

Test When Killing Two Civs on One Turn

In this test, two civs were destroyed on the same turn. A save was created later in the same game used in the first test above and 100 new trials were done. For the first 50, the Spanish were destroyed first, followed by the Vikings. For the second 50, the Vikings were destroyed first, then the Spanish. The order of execution was changed to see if it made any difference. The Spanish were the original yellow civ in the game. The Vikings were the first respawn of the Germans, since the Germans had been killed by the Japanese earlier in the game.

nr = no respawn

For this 50, yellow first, then blue

1 nr; French(9,75)
2 nr; nr
3 nr; nr
4 Aztecs(12,66); nr
5 nr; French(8,72)
6 nr; nr
7 nr; nr
8 nr; nr
9 nr; nr
10 nr; nr
11 nr; French(10,64)
12 nr; nr
13 nr; nr
14 Aztecs(70,54); nr
15 Aztecs(11,65); nr
16 nr; nr
17 Aztecs(12,64); nr
18 nr; nr
19 nr; nr
20 nr; nr
21 nr; French(8,74)
22 nr; nr
23 Egyptians(9,69); nr
24 nr; nr
25 Aztecs(16,34); nr
26 Aztecs(9,69); nr
27 nr; nr
28 nr; nr
29 nr; nr
30 nr; French(10,66)
31 Egyptians(12,64); French(9,73)
32 nr; nr
33 nr; nr
34 Egyptians(12,64); nr
35 nr; nr
36 nr; nr
37 nr; nr
38 nr; nr
39 nr; nr
40 Aztecs(12,64); French(9,69)
41 Aztecs(11,65); nr
42 nr; nr
43 Aztecs(19,33); nr
44 nr; nr
45 Aztecs(7,75); nr
46 Aztecs(8,74); French(16,34)
47 nr; nr
48 nr; nr
49 nr; nr
50 Egyptians(10,68); nr

Tabulated results were:


For Yellow only:

nr – 35 times (70%)
Aztecs – 11 times (22%)
Egyptians – 4 times (8%)

The chance of a spawn was 30%


For Blue only:

nr – 42 times (84%)
French – 8 times (16%)

Here the chance of a respawn was only 16%, which helps confirm the results of the previous test, where odds diminish for successive respawns.


Combined Results:

nr – 77 times (77%)
spawn occurs – 23 times (23%)



For this 50, Blue was killed first, then yellow:

1 nr; Egyptians(9,69)
2 nr; nr
3 French(12,64); Egyptians(13,65) they are next to each other and declare war!
4 nr; nr
5 French(8,76); nr
6 nr; nr
7 nr; nr
8 nr; nr
9 French(12,64); nr
10 French(7,65); Egyptians(8,64)
11 nr; nr
12 French(9,71); nr
13 nr; nr
14 nr; nr
15 nr; nr
16 nr; nr
17 nr; nr
18 nr; Aztecs(8,70)
19 nr; nr
20 nr; nr
21 nr; nr
22 nr; nr
23 nr; Aztecs(8,64)
24 nr; nr
25 nr; nr
26 nr; Egyptians(12,64)
27 nr; nr
28 nr; nr
29 nr; nr
30 nr; nr
31 nr; nr
32 French(11,65); nr
33 French(13,65); nr
34 French(19,33); nr
35 nr; nr
36 nr; nr
37 nr; Egyptians(8,76)
38 nr; nr
39 nr; Aztecs(12,64)
40 nr; Aztecs(12,66)
41 nr; Aztecs(9,75)
42 nr; nr
43 French(12,66); nr
44 nr; nr
45 nr; nr
46 nr; nr
47 nr; nr
48 French(10,66); Aztecs(12,64)
49 nr; nr
50 nr; nr

Results were:

For Blue only:

nr – 40 times (80%)
French – 10 times (20%)


For Yellow only:

nr – 39 times (78%)
Aztecs – 6 times (12%)
Egyptians – 5 times (10%)

Spawns occurred for yellow 22% of the time, about the same as for blue, but they are still more likely for yellow’s first spawn than for blue’s second.


Combined Results:

nr – 79 times (79%)
spawn occurs – 21 times (21%)

Little difference here between the two sets.


Finally, when the results for all 100 trials are tabulated together:

For Yellow:

nr – 74 times (74%)
spawns – 26 times (26%)

For Blue:

nr – 82 times (82%)
spawns – 18 times (18%)


The distribution of spawned settlers was similar in many respects to those in the first test, with 95 of them located in the area where 96 turned up in the previous test. Again, the distributions were tabulated:

Main area

(12,64) – 9 times
(9,69) – 4 times
(12,66) – 3 times
(11,65) – 3 times
(9,75) - 2 times
(8,74) - 2 times
(10,66) - 2 times
(13,65) - 2 times
(8,76) - 2 times
(8,64) - 2 times

Minor area

(16,34) – 2 times
(19,33) – 2 times

Indeed, not much had changed in the map, although a new Chinese city had been added to the east. It seems that this new city pushed the main respawn area a bit west, changing the distribution of spawn sites, as can be seen in the screen shot accompanying this post. Of the remaining 5 respawns, 4 were clustered near each other again, but in a new area, perhaps reflecting changes that had occurred as cities were built and/or destroyed as the game progressed. The remaining respawn was in a unique spot.

The consistency found in respawn locations during this game is remarkable. One might think that multiple kills on one turn might end up with civs respawning all over the map, but this was not the case in this test. When yellow and blue both respawned on the same turn, they were both most likely to be in that favorite respawn area. In fact, once they were placed side by side, and as is typical with the AI, hostilities commenced immediately!

From these limited tests, it would seem that areas far from, but surrounded by existing cities are favored as respawning sites by the game.

(Note: In the screen shot, only one color was used to represent new settlers, since yellow and blue would have shared many spots)

[La Fayette]

great job, solo!

1) You justify the title of the thread: "...how it works"

2) Your results will be very useful to anyone willing to dig further.

3) I am not surprised at all if there is a random factor, because I am convinced that there is also the same kind of randomization at work when a dip/spy destroys improvements (usually city walls last, but not always).

[Zenon]

It is good to see more findings on the respawn issue. Good test, Solo!

After a break I couldn´t resist to play the new tournament. I just destroyed the Japanese in 40 AD and had a no respawn. My former tests showed that the random factor for respawns depends on number of civs left, number of turns played ( with an end date at 1500 AD on deity discovered by monk), number of respawns of the same colour and difficulty level.

The place were respawns appear seems to depend on proximity to the civ destroyed and proximity to the next city. A random factor is at work here, too.

A very complex issue, indeed.

Zenon

[solo]

Good to see you around, Zenon!

I agree with you about the types of factors involved, and that it would take many more additional tests to see how each contributes towards the results. It's probably not worth the effort figuring out the exact proportion of chances. I've surmised enough from my own limited tests to answer my own questions about respawns.

I look forward to seeing your result in this game.

[Peaster]

This un-interesting post has been erased (sorry).

[Peaster]

IMHO, this is far from settled. Some reports from GOTM46 [WARNING SPOILER INFO AHEAD]-

LaFayette reports 20 respawns. (MPGE?)
Zenon's report is similar. (Civ2 2.42?)
But I had only 1 respawn. (MPGE)

I reloaded the final position (180AD) just before I killed the purple civ. I tried 5 times and never got a re-spawn. I backed up to my 160AD save and replayed the last two turns (did this twice), still with no re-spawn.

Solo's work proves there is a random factor, but there is clearly a non-random factor, too.

GOTM46 Map data: Americans, Deity, small map, 5 civs, restless tribes, Civ2/MPGE. Ended with 23 cities in 180AD.

BTW - When respawns do occur, I think the civ appearing can repeat unpredictably, though I didn't notice this happening in Solo's report. For example:

Start = Mongol. Then Sioux -> Mongol, etc . (The Indians can be skipped).

[Zenon]

Peaster,

Yes, I played 2,42. I think there were 2 factors at work in the current GOTM:
1. Respawns are more likely in BC years.
2. It would be worth to compare how many cities were on the map. In my game most of the AI cities were destroyed leaving good respawn areas.

The save you wanted is attached. For GOTM players: my game is out of competition but still a spoiler for the current game.

[Peaster]

@zenon: I loaded your game (nice one BTW) 15 times and only got 3 re-spawns, a MUCH rate lower than what you reported in your game. It would be interesting to see what kind of re-spawn rate you get from replaying this position yourself. I wonder if there is some factor which is not even part of the *.sav file, which affects re-spawns.

BTW - I have had miserable games like yours and LaFayette's, with 15-20 respawns. Just not recently.

You had 16 cities to my 23. The 3 respawns I got from your position were all near to where the last English settlers were killed. This was about 15 squares from the nearest American city. [In my game I had cities within about 8 squares of all decent land].

I also did some more tests today using my own game. I used a 1ad save this time, just in case there's some glitch when you replay the final turn of a game. I tried capturing/razing the last Mongol city playing the Americans and then playing the Japanese (using cheat mode). I got only 4 respawns out of 25 tries. The highest rate came playing the Japs, and razing the city, allowing 3/10 respawns (still far below the zenon/LaF rate).

[solo]

I thought I would test a little more to provide more input.

I also downloaded Zenon's save and got 6 respawns in 20 attempts using 2.42, with one string of 8 attempts in a row without a respawn. The first 4 respawns came within the first 7 attempts.

I think there may be something odd going on in the game, some sort of bug, one indicator of this being that on attempts that succeed, the message is that the English were destroyed by barbs. Also twice during my test when I killed a respawn immediately, another settler appeared on the map, moved and after clicking for the next turn I got the end of game display. After this I kept playing and killed this new settler and "won" the game again!

Peaster, are you using MGE? There could be a difference between game versions, too.

Another thing that might affect tests is whether or not a complete exit is made from the game before each trial. In my tests I did not exit, and this might have tainted the results in some unknown way.

[Peaster]

solo: Yes, I have MGE. [remember that LaF got a lot of re-spawns, and I think he has MGE, too]. I did not restart the game between tests. I did not let zenon's 50BC turns end before reloading b/c I didn't want to sit thru the `guillotine routine'.

I also noticed some patterns like yours. In my test of zenon's save, the results were YYNNN NNNNY NNNNN (Y means "Yes, there was a respawn").

In my official GOTM46, the pattern was YNNNNN, which may just be a coincidence.

I replayed my 180AD kill a few times and then loaded my 160AD save. At the end of the 160AD turn I had not conquered, but I got the victory messages anyway and had to play 160AD again to clear it. This is one reason I think the program stores data in some file/memory besides the *.sav file.

This GOTM46 story seemed familiar from somewhere, so I looked back at GOTM45 and found these notes in the spoiler -

LaFayette " But respawning was VERY active in my game (probably more than 20 respawns )"

Peaster "Luckily, the first 5 civs did not respawn. The 6th did, twice."

Let me emphasize - I HAVE had some training games with 15-20 respawns, like zenon/LaF, about 1-2 months ago. I do not recall any games with a medium number of respawns (like 8-10)(not sure).

Just speculating - but it seems possible that variables stored from a previous game could affect the next ones, especially if the games are loaded from saves rather than freshly created. I may have started GOTM46 with a big advantage without even knowing it! That advantage may still be in my wires somewhere.


Update [11/8] : In the interest of civ2 science, I started a fresh game to "clear my system" and repeated the test on zenon's save, getting

NYNNN NYYNN YYNNN = 33% = about twice the 15% rate I got in the previous 40 trials. I started another game, did it again, and got

NNNNN NNNNN NNNNN = 0%. To me, these indicate
your rate persists until you start a new game. Maybe the varying restart rate was intended for fairness - kind of like the number of starting techs. Of course, this would be more convincing if I had gotten an 80% rate at least once, or if I had the patience to do 100 trials at a time like solo.

12/2/04 Maybe I was wrong. In the Holiday tournament, I had about 7-8 respawns (60%). After the game, I tried zenon's save again, and got only 2/15.

[SlowThinker]

Are there any conclusion that may help me where to expect an AI restart?
Is it more likely on the original home continents?
I noticed on solo's maps that origination of Settlers is concentraced to specific area, but how is this area choosed?

[solo]

ST,

I think the locations depend on the distribution of existing cities, and that respawns are usually found in areas equally remote from existing cities. For each respawn, I believe several sites are identified as candidates and each of these is given a probability of being chosen based on how suitable (i.e. remote) it is. Whenever cities are added to or razed from the map, the likeliest respawn area(s) almost always change.