Page 2 of 2

Re: Wood Duck > Big 4? A simulation-based tier list

Posted: 14 November 2026, 22:40
by horse_feed
yeah good idea I'll post one in a few days

Re: Wood Duck > Big 4? A simulation-based tier list

Posted: 17 November 2026, 05:44
by crocuspocus
I'm a bit worried about the random bird stuff, since "how good is this bird when I'm randomly forced to play it" is not really the question most people have in mind when they compare how good different birds are. Could you tell us for, say, the 50th percentile bird (in terms of how much the computer likes playing it) what fraction of its plays were voluntary vs forced? How about for the 30th percentile bird? It would be good if plays for these birds were mostly voluntary (so the random playing practically only matters for the very unpopular birds).

Re: Wood Duck > Big 4? A simulation-based tier list

Posted: 17 November 2026, 05:47
by crocuspocus
Also just a sanity check -- where is the computer putting the power birds? If it's not the meadow that might explain some things...

Re: Wood Duck > Big 4? A simulation-based tier list

Posted: 17 November 2026, 06:56
by FrankJones
I find it truly bizarre that in an expansion in which there are 15 Nectar points up for grabs (Well, 21, but 15 is the max any one player can get), and a raven can take 2 from the supply while activating a non-forest habitat, that people still say, "Oh, the ravens aren't all that good in Oceania, because .... reasons."

Especially when the sometimes-difficult-to-play raven becomes so easily playable in Oceania with 6 starting food including one nectar, meaning it can be played with the nectar and any 2 food While still being possible to keep one other card and one other food.

WHAT.

Re: Wood Duck > Big 4? A simulation-based tier list

Posted: 17 November 2026, 19:32
by horse_feed
crocuspocus wrote: ↑17 November 2026, 05:44 I'm a bit worried about the random bird stuff, since "how good is this bird when I'm randomly forced to play it" is not really the question most people have in mind when they compare how good different birds are. Could you tell us for, say, the 50th percentile bird (in terms of how much the computer likes playing it) what fraction of its plays were voluntary vs forced? How about for the 30th percentile bird? It would be good if plays for these birds were mostly voluntary (so the random playing practically only matters for the very unpopular birds).
Forced plays make up a very small percentage of each bird's total plays. This procedure was introduced only to get a bit more information on birds which the AI was truly reluctant to ever play, such as the Acorn Woodpecker or the Long-Tailed Tit.

Here's some numbers for play rate percentile vs. proportion of plays which were forced:

Code: Select all

Popularity percentile     Forced play rate
(by how often played)    
------------------------------------------------------
 0-10  (least played)     17.4%
10-20                     10.8%
20-30                      6.0%
30-40                      3.4% 
40-50                      2.7%
50-60                      2.0%
60-70                      1.6%
70-80                      1.4% 
80-90                      1.3%
90-100 (most played)       1.0% 
------------------------------------------------------
All birds                  1.8%             1.7%
So each bird's score is primarily informed by cases where the AI deliberately chose to play the bird, and for most birds, this accounts for the vast majority of cases.
crocuspocus wrote: ↑17 November 2026, 05:47 Also just a sanity check -- where is the computer putting the power birds? If it's not the meadow that might explain some things...
Grassland, pretty much always:

Code: Select all

Bird               Plays   In grassland
----------------------------------------
Killdeer            203     99%
Franklin's Gull     230     96%
Chihuahuan Raven     86    100% (its only habitat)
Common Raven         89     82%
American Crow       219     84%

Re: Wood Duck > Big 4? A simulation-based tier list

Posted: 17 November 2026, 19:57
by horse_feed
FrankJones wrote: ↑17 November 2026, 06:56 I find it truly bizarre that in an expansion in which there are 15 Nectar points up for grabs (Well, 21, but 15 is the max any one player can get), and a raven can take 2 from the supply while activating a non-forest habitat, that people still say, "Oh, the ravens aren't all that good in Oceania, because .... reasons."

Especially when the sometimes-difficult-to-play raven becomes so easily playable in Oceania with 6 starting food including one nectar, meaning it can be played with the nectar and any 2 food While still being possible to keep one other card and one other food.

WHAT.
I personally don't think Common/Chihuahuan Raven are S or even A in Oceania anymore:
  • Eggs are scarcer in grassland activations.
  • Food is cheaper to get from forest activations.
  • Availability of nectar reduces the value of being able to get arbitrary kinds of food from supply.
  • The expansions add a lot of birds that get you food in other ways.
There's just much more opportunity cost than there used to be with pursuing a grassland engine. For this bot, brown birds that live only in the grassland are the weakest habitat–color combo in the list, and a forced grassland activation is the worst of the four actions in every round before the last.

That said, I'll admit C/D for the ravens seems really low. The computer plays a raven in about 1 game in 200 and almost never takes the egg trade when it does. Perhaps the computer knows the game better than most humans, or maybe it just hasn't properly learned the bird. I'm going to do some more experiments to probe this further.

Re: Wood Duck > Big 4? A simulation-based tier list

Posted: 17 November 2026, 20:57
by FrankJones
Are eggs really that much more scarce? Once 3 birds are in the grassland, extra cards can be converted into eggs, which is not possible in the base game player mat. Base game, if you need to draw cards several times because you're getting trash cards, too bad. They have little or no use without a "tuck and lay egg" bird. And, in Oceania, it's easy to have more cards since cards are easier to get.

Sure, food is easier to get, but being able to take 2 nectar from the supply while taking an inherently points-gaining action is still very strong. in a 2-player game, there might not be nectar in the feeder, and being able to avoid the feeder while the other player is stuck dealing with it gives a big advantage.

I'm not sure I follow your point about reduced value of getting arbitrary food from the supply. The food gained from the raven will mostly be nectar. And there are a lot of nectar points up for grabs.

Re: Wood Duck > Big 4? A simulation-based tier list

Posted: 17 November 2026, 22:54
by RazorOz
The Ravens are far more busted in Oceania imo, a round 1 raven basically guarantees you win all the nectar bonuses unless someone else has the same engine going. That's a free 15 points by itself, and it's worse than that because it's 15 points you're getting at the expense of your opponents. There's still no benefit in Oceania of doing the gain food action if you don't need to, and it's clearly better to not have to be doing it.

What I'd be interested in here is what level is AI playing at, ie. can it beat anyone of substance at the game, because if not, the data itself is not that useful. I see AI analysis being done for a lot of games recently, but if the AI itself is not playing very well the data is not very useful.

Re: Wood Duck > Big 4? A simulation-based tier list

Posted: 17 November 2026, 23:51
by horse_feed
I'm going to play my bots now if anyone wants to watch: https://www.twitch.tv/horse_feed

will post link to vod once I go offline