Season 19 Meta Data for 3p Prelude Arena

[Glados93]

Very interesting data, with quite a bit of things coming as a shock to me (some of my favorite preludes are providing middle of the pack results).

Some thoughts and questions:

1 - Where I usually play, the perception is that the top tier corps are Point Luna and Vitor (+ Manutech and Poseidon at 4p). Evidently, I play with Venus and Colonies added (sometimes Turmoil, but seldomly). Would you say that opinion is simply wrong or is it expected that they get a big boost in win rate when Venus/Colonies are added? Would you be willing to take a swing as to how the entire corps power rankings shift when those expansions are added?

2 - IC has a high win rate, but is one of the least picked corps. I'm guessing it means it's not considered strong by the masses, and thus said masses don't drop its WR by playing it? Or is it that it is just picked when perfect conditions appear, which boosts its WR? I tend to like it very much with Media Group and Optimal Aerobreaking, or Advanced Alloys or when I can immediately play 40mc worth of building cards that give me good production, but that is not that common.

3 - Ecology Experts has a bad WR (23rd prelude out of 35), while Kelp Farming and Trees are among the best gen 1 cards (respectively #6 and #12 out of 158). Meanwhile, Fish is #131. What do you conclude based on that? High WR when it's paired with the right cards, bad otherwise? Or just bad? I reckon gen 1 Kelp Farming might be a result of 6 oceans (or 4 oceans + Inventrix/AT/SD), but gen 1 Trees are probably always paired with EE. If you get EE with Kelp/Trees and see an adversary with Ecoline, would you pivot to Birds instead?

4 - Any chance we can get stats on how often cards are consciently picked rather than randomly drawn? Some events are probably only played because they were drawn for free and you have Media Group, which speaks more about Media Group (and card draw engines) than the event itself.

Yeah the data is quite interesting with a mix of expected and surprising results! We should however remember to reserve our conclusions for when i have looked at more high level games to get a larger sample size.

Our of curiosity, which favorite preludes are you referring to?

I will try to answer your questions. Disclaimer: Im a new TfM player and have not been part of the "Old Guard" of players on Steam. I base my opinions mostly on data and what i have heard from players with more experience than me. I play exclusively 3p, which biases my ideas for this format.

1. Point Luna (PL), Vitor, Manutech and Poseidon are all considered top tier corps by the majority of high level players. However, in base + prelude, where the dominant strategy is usually terraforming and ground game control, the real S tier corps are Credior and Arcadian Communities (AC). PL, Vitor and Manutech are also extremely strong here, but not quite as flexible as Credicor and AC. Ecoline also deserves a mention in base+prelude as an extremely strong, albeit one-dimensional corp with the right hand. Point Luna is very high-rolly (a bit like AI-central). If you hit the draws it can go super-saiyan straight to the top of S-tier, but in most games you dont hit all the right draws of the bat and the power level is a bit lower than Credicor and AC. When you add expansions (especially Colonies), the game win condition leans much more into engine setup. Here Manutech becomes extremely strong (to the point of busted) along with Poseidon, Vitor and PL becoming stronger. Credicor remains strong, while Ecoline becomes much worse.

2. Interplanetary Cinematics (IC) has a very high winrate in the current data, but a lower playrate than Credicor. As you reason, this is because it is mostly picked when the conditions are favorable. i.e Media Group or Alloys in starting hand, at which point its extremely strong. But I think the corp might also be undervalued by most players. Its scoring extremely well in the data both for the current base +prelude format, but it also turned out as the number 2 corp in the previous season of 1500+ games (base game). Additionally, it has one of the highest wins above predicted (WAP) in the data. This metric evaluates the winning ability of the corps based on the elo of the players in each game. The high WAP indicates that IC is an upset corp, where lower rated players tend to beat much higher level players when playing it. This could possibly be explained by IC nudging the lower-rated players to play events and rush the game, which is typically a strong strategy. So you could say it nudges players to play better lines than they would have with other corps. The same can be said about Tharis Republic, which also has high WAP. Lower-rated players tend to spam cities, which is typically a very bad strategy, but with Tharsis you are rewarding for doing so. This means Tharsis improves the lines lower-rated players tend to choose, and thus they end up winning more than they usually would have.

3. I had the same thoughts when i first saw Ecology Experts (EE) in the data. I dug into it a bit, and the low winrate is definitely explained by players using it on cards like Fish. When used on Trees and Kelp it has a winrate around 50%, which puts it comfortably into top tier, if not the best prelude. Ecoline with EE on Kelp currently has 100% winrate. About your specific questions: If you can hit Ecoline with EE on Birds i think you should definitely do so in 2p, you should strongly consider it in 3p unless the third player is much higher-skill level than Ecoline, and in 4p its probably not optimal.

4. Sadly, getting info on when cards are drafted / seen and not only played is a lot more cumbersome. The replay logs record from the perspective of a single player, and as such you need to access the replay log for each player individually to get this info. There is a limit on the number of replay logs you can view per day, so this would cut the sample size you can view per day by 1/3 for 3p games. Its possible, but would just take a while.

[Glados93]

Honestly, fun data to look at but mostly meaningless. I do really appreciatie the effort put in by the author.

The lack of context along with the format being a hodgepodge of top players frequently playing against players that are 400-500 elo below them (this equates to an average winrate of 5-7% for the lowerrated player) makes the data borderline useless in a competitive context.

Most good players that have any form of knowledge on card picking and efficient will simply crush 95% of the field, irregardless of the cards they play. A great example of this is the ranking of Tharsis/Saturn which are both influenced by opponents playing into them by playing unncessary cities or jovs.

The "good" cards are heavily skewed towards late game cards that provide winmore value. There is no pick rate (how often did a player take ice cap melting in gen 1/2/3 etc) available. The opposite is true for card ls like nuclear power, carbonate processing and geothermal power which are all bottomtier according to this list.

Directly extrapolating information about how to play from this dataset is, imho, unwise and also not based in any form of real analysis.

I agree with most of your statements. However, while the data may be mostly useless to you as a high-level player, it can still be useful for newer or less experienced players. For example, the data clearly supports that certain cards that are passed around way too much in draft are very strong, and should be considered with more care. Here im specifically thinking of NRA, Insects and Indentured, which have very high playrate, so are not only played by higher level players, but maintain high winrates. If you are surprised to see a certain card with many plays and high winrate among the top rankings, then you should reconsider your thoughts on the card. This is mostly useful for newer players, that have not had the experience to know which cards these are.

As to your point of top players winning most games with any cards; this is true, but the data also records Wins Above Predicted (WAP). I.e how often did lower rated players get an upset and beat high level players. It can be informative to look into which cards made this happen. The opposite is also true, which cards where typically played by higher rated players when they lost? Here GRASS is an example. This card has somewhat high winrate at rank 83 overall, but has a negative WAP. I.e. this card was played by many top level players, but actually negatively impacted their base expected winrate. This is an indication that some high-level players (in Arena) overvalue the card.

The actual dataset records the elo of every player, so more information could be extracted if someone wanted to. I.e. look at elo bins to see which cards where played and lead to which results in each bin.

As to your last point - absolutely agree. You cannot directly extrapolate how to play from this data. The strength of any card in TfM is heavily dependent on the current game-state. However, some cards a just generally stronger than others, and this knowledge can be useful for newer players.

[AdrianoFreitas]

Our of curiosity, which favorite preludes are you referring to?

I was mainly referring to the ones that give 6 mc production worth of stuff (Business Empire, Galilean Mining and Metals Company). I also like Research Network and Power Generation (with Colonies).

[Dikvarkentje]

Honestly, fun data to look at but mostly meaningless. I do really appreciatie the effort put in by the author.

The lack of context along with the format being a hodgepodge of top players frequently playing against players that are 400-500 elo below them (this equates to an average winrate of 5-7% for the lowerrated player) makes the data borderline useless in a competitive context.

Most good players that have any form of knowledge on card picking and efficient will simply crush 95% of the field, irregardless of the cards they play. A great example of this is the ranking of Tharsis/Saturn which are both influenced by opponents playing into them by playing unncessary cities or jovs.

The "good" cards are heavily skewed towards late game cards that provide winmore value. There is no pick rate (how often did a player take ice cap melting in gen 1/2/3 etc) available. The opposite is true for card ls like nuclear power, carbonate processing and geothermal power which are all bottomtier according to this list.

Directly extrapolating information about how to play from this dataset is, imho, unwise and also not based in any form of real analysis.

I agree with most of your statements. However, while the data may be mostly useless to you as a high-level player, it can still be useful for newer or less experienced players. For example, the data clearly supports that certain cards that are passed around way too much in draft are very strong, and should be considered with more care. Here im specifically thinking of NRA, Insects and Indentured, which have very high playrate, so are not only played by higher level players, but maintain high winrates. If you are surprised to see a certain card with many plays and high winrate among the top rankings, then you should reconsider your thoughts on the card. This is mostly useful for newer players, that have not had the experience to know which cards these are.

As to your point of top players winning most games with any cards; this is true, but the data also records Wins Above Predicted (WAP). I.e how often did lower rated players get an upset and beat high level players. It can be informative to look into which cards made this happen. The opposite is also true, which cards where typically played by higher rated players when they lost? Here GRASS is an example. This card has somewhat high winrate at rank 83 overall, but has a negative WAP. I.e. this card was played by many top level players, but actually negatively impacted their base expected winrate. This is an indication that some high-level players (in Arena) overvalue the card.

The actual dataset records the elo of every player, so more information could be extracted if someone wanted to. I.e. look at elo bins to see which cards where played and lead to which results in each bin.

As to your last point - absolutely agree. You cannot directly extrapolate how to play from this data. The strength of any card in TfM is heavily dependent on the current game-state. However, some cards a just generally stronger than others, and this knowledge can be useful for newer players.

Thank you for your thoughtful and thorough reply. My apologies if my original response came off as offensive, it was not meant that way (however reading back it seems a little harsh/rude).

If you don't mind explaining, how are you calculating the WAP? Is it a simple X < or > Y or is there a weighting involved?

[Glados93]

No worries! I completely understand how the data mainly highlights conclusions that seem obvious to you. This is as much a testament to your experience in the game as it is to the data supporting what you already know to be true

Here is how WAP is calculated:
Wins Above Predicted (WAP) is an Elo-corrected measure of strength, reflecting the performance of a specific card or corporation, adjusted for the Elo ratings of players in each game where that card or corp was played. A higher positive WAP score indicates stronger performance.
For 3-player games, I calculate the WAP score for each card or corp as follows:

WAP = mean_over_games(actual_points_game - predicted_points_game),

where
* mean_over_games is a mean over all games where the card/corp was played
* actual_points_game are win points assigned as 2 for 1st place, 1 for 2nd place, and 0 for 3rd place.
* predicted_points_game represent the expected points based on each player's Elo, calculated according to the Elo ratings of their two opponents.

For example
* if a low Elo player wins against two high Elo players, where the win probability was 14% and 13%, their corp would get a WAP = (2 - (0.14+0.13)) = 1.73 for that game.
* If a high Elo player won against two low Elo players with 95% and 95% win probability based on elo, their coporation would get a WAP of (2 - (0.95+0.95)) = 0.1 for that game.

Thus, wins that are less expected result in higher WAP scores. The WAP score can be understood as the average points a corporation or card achieves above what is expected given the players’ Elo ratings. If a corporation has a negative WAP score, it means it typically underperforms relative to expectations based on player Elo; in other words, high Elo players tend to lose more often when using this corporation compared to those using corporations with higher WAP scores.

The WAP score, as calculated, can be interpreted directly as the mean expected change in win points when playing the card. For example, if a card has a WAP score of 0.1, this implies a 0.1 win point (10%) increase in the player's winning probability, added to their base win probability (as predicted by Elo).

As a caveat, remember that the overall sample size in the data is still fairly small (about 1,500 games), and that these WAP scores only look in specific games where a card/corp appeared. For this reason we should wait for a larger sample size before we can draw any real conclusion based on the computed WAP scores.

[RoyalRook]

jovs is not efficient, but definitely need to keep an eye on. Especially in a live game and you can really catch someone sleeping with it. Most game I lost is b/c of jovs. Cards like Miranda, looks so innocent but can really jump start an engine especially with Saturn, Methane from Titan feels like such an impactful card. I am more scared of someone drop titan on gen 1 than IO. Maybe that's why an early miranda with couple earth tag support was able to rank so high, by unlocking more jov options.