10km twin rivers ai vs ai tournament results now up:
https://forum.faforever.com/topic/9755/ai-vs-ai-tournament-series

AI tourney: 10km Twin Rivers
Since the vote on the next map for the tournament was tied between twin rivers and white fire, I decided to go with Twin Rivers.

Each AI faces off against M28AI in a 1v1 on Twin Rivers to see what AIx build and resource cheat multiplier it needs to win half of its games.

Setup
M28AI will fight against the enemy AIx using mirrored factions. Based on how heavy the win/loss the AI modifier gets adjusted to try and quickly arrive at the AIx modifier that means half of games get lost and half get won; in some cases where significant factional differences are suspected or I have the patience to run more games, games will be run with each faction to identify the lowest modifier required with that faction for the AI to win. Results will then be displayed for the estimated modifier (e.g. if an AI wins with a 1.2 modifier with 3 factions, but a 1.1 modifier with 1, then it is assumed the 3 factions would draw with a 1.15 modifier, and the other faction would draw with a 1.05 modifier, meaning it's shown with a modifier of 1.125, i.e. (1.15*3+1.05)/4.

Changes to AI
Since the last tournament there have been minor updates to M28AI and RNGAI, and more interestingly a new AI mod is on the FAF vault, currently on v102, called Vanguard AI.
See the 5km tournament for details on other changes made in the previous 3 years
The results
Summary

Detailed results
The full results including the data used for the above graphs can be seen in the Data worksheet of the following file:

https://docs.google.com/spreadsheets/d/101PqVFKozMaWcHbaceIbabLGDsfdnoY-/edit?usp=sharing&ouid=100973959280546778272&rtpof=true&sd=true

Other Observations

• M28 was generally hurt by going for navy on one of the ponds, with a significant mass investment for minimal benefit
• AIs with high modifiers would typically win with nukes or an early experimental
• M28 was able to win and hold both plateaus in almost every game it won
• Land played a significant role with air generally not having much impact unless an early air experimental from a high AIx modifier, or M28 using gunships defensively
• Swarm's poor performance (most AI would be expected to require a lower AIx modifier the larger the map) appeared due to not building/using anti-air units effectively until late in the game allowing M28's gunships to attack its tanks with impunity
• Games on LOUD would generally go on far longer than on FAF (despite it being a 10km map which I thought might be enough to mean shorter games - I was curious whether LOUD's typically long game times are due primarily to its preference for 20-40km map sizes). My guess is a significant part of the reason for the long game times is the effectiveness of shielding (e.g. no shield damage overflow and rapid recharge times meant M28 in one game built more than 8 T3 artillery before it could break through T3 shields, in another game a mavor and 3 T3 arti were not sufficient for it to break through)
• I've updated the 5km tournament results to include the new Vanguard AI, and also to re-run all the steam results in case they were affected by a subsequent bug discovered that could cause AIx overwhelm settings to be enabled and adjust resource multipliers (however the steam results only changed marginally so the bug didn't appear to have had a significant impact)
• The average AIx modifier on twin rivers was 1.57, compared with 1.9 for Theta Passage. Such a decrease is expected since as the map size increases ACUs have less impact, it takes longer for land units to reach the enemy (so greater risk of being defended and/or out-teched by the units they meet), a greater proportion of mexes tend to be outside the core base (meaning greater benefit from the initial expansion boost provided by a build rate modifier), the economy (and hence resource modifier) generally plays a bigger role due to this, and high impact units that can end the game without significant investment to defend against them (like a land experimental or a nuke built with a build rate modifier) become much easier to build (more resources, and more time).

v322 Update
32 changes, including:

• Asfs should do a better job at protecting game-enders from enemy strat snipe attempts
• ACUs should be more likely to run if they represent only a small amount of the total allied or enemy threat (due to the risks of being easily focused down)
• Reduced naval factory assistance in low value ponds
• Various late-game changes, including optimising an issue that could cause the game to pause for half a minute when M28 started building a game-ender
• Various LOUD specific changes

Acknowledgements

• Relent0r - replay vs M28Easy (following which I made M28Easy less likely to build sniperbots since it doesnt kite with them)

v321 Update
16 changes based on a Radde 5km land replay, including:

• ACU should be more aggressive when injured if it has a personal shield
• Land scouts should stay closer to ACUs so theyre less likely to fight an enemy ACU they cant see
• Once M28 reaches T2 land it should be even less likely to build T1 land units (or more t1 land factories)
• When looking for nearby enemies to run from, zones adjacent to the current zone should be considered in some cases (let me know if games are significantly slower following this change)

Acknowledgements

• Radde - 5km land mapgen replay beating 1.5/1.0 M28

v320 Update
16 changes, mostly relating to small naval maps, including:

• ACU should help build the first naval factory and initial naval units (on small naval maps)
• Inties should be built if the reason M28 isn't attacking with torpedo bombers is the enemy having too many nearby air units
• Fixed a bug relating to MAA introduced in v318
• M28 should be less likely to pause it's main T1 naval factory when stalling energy

Acknowledgements

• Radde - mapgen replay beating 1.6 M28 AIx

Trophy award

• Radde - Sladow trophy - beating 1.6 resource and build rate M28 (on a 5km mapgen)

Some maps and games are more likely to flame than others.
My advice - stay clear of setons and dual gap, people get very angry over small things and/or not playing the way they expect.

3v3 matchmaker is usually fairly chill (but there are exceptions), partly because there's less of a pre-set meta
Astro craters maps used to be relatively chill as well but it's been a few years since I've played them

there are occasionally team games vs AI which are fairly rare to have flaming

inevitably you'll come across some people who are toxic, but it's definitely a relatively small minority of games I play (I usually play 3v3 TMM). There's also 1v1 where it's less likely since there's no teammate to complain about how you play.

v319 Hotfix
Mini update fixing a bug that meant M28's engineer logic wouldn't work on a mapgen that featured a tiny corner pond

Thanks to Radde for highlighting the error

v318 Update
Various changes focused on further optimisations to achieve some marginal performance gains, in particular rewriting how M28 calculates the threat of units to use a quicker system of referencing (at the cost of being a bit less readable and with a greater risk of an invalid threat reference being used when I'm drafting future code), along with some more minor changes to several other functions.

Basic tests suggest the changes might have improved game speed by a few percent (but the method of estimating this is very imprecise).

I also experimented with switching how M28 works out if it can build somewhere, to move most of the complicated logic to only run when selecting the best location available (instead of calculating in advance), but initial results suggested either no improvement or a negative improvement, so I abandoned this approach.

v317 Update
28 changes, almost entirely relating to naval maps, including:

• Fixed an issue that led to enemy Seraphim destroyers being double-counted for threat calculations (once as a sub, once as a destroyer)
• Improved logic for when naval units decide to attack an enemy, and then some of the attacking force moves into a water zone closer to the enemy (previously this would lead e.g. M28 having 2 destroys vs 1, but retreating both destroyers)
• Seraphim destroyers should no longer try and dodge enemy shots while firing (as this led to them cancelling their attack, and so caused a major DPS reduction if the enemy had a cruiser)
• Adjusted how subs handle torpedo launchers (to e.g. make it less likely they attack-move an enemy naval factory while taking fire from a torpedo launcher)
• Various adjustments to decisions on whether M28 should attack the enemy with its navy when it doesn't outrange them
• If M28 has a naval factory and is low on mass its land factories should be less likely to build hover units and amphibious units in response to a nearby enemy naval threat, so that more mass can go into navy
• M28's naval units early on should hunt for a potential enemy naval factory (previously they'd assume there was none, due to lack of intel, meaning M28 could establish an early naval advantage and then waste it by leaving its units idle)
• Made a (blind) change to Novax targeting logic so M28 assumes it takes at least 1s to kill a unit, given the recent nerf to novax (which I'm assuming means that the novax can no longer switch targets instantly)

Acknowledgements

• Radde - Setons-like mapgen replay that formed the basis for majority of changes made

v315-316 Update

• SCTA compatibility added - M28 SCTA ARM and M28 SCTA CORE can now be selected as an option with the SCTA mod. If M28 is enabled then M28 will then play with the relevant SCTA faction. It's strongly recommended you also enable the SCTA balance mod if using M28 with SCTA.
• Naval units such as frigates should be slightly better at kiting auroras
• Fixed a bug with building MAA to support a land experimental
• Several other bugfixes, including removing debugging options that were accidentality left in v315

Acknowledgements

• Dragun101 - Helping with implementing M28 SCTA compatibility, in particular by allowing M28 to be selected as an AI option via the SCTA mod
• Radde - 2 mapgen replays
• Fearghal - mentioning v315 release was adjusting game speed (where it was set to adjustable)

v314 Update

• ACUs should be more likely to start a second upgrade of a similar energy cost to a recently completed upgrade despite not overflowing energy
• Fixed a bug causing ACUs to stay stuck in their base idling
• Fixed a bug that meant air support factories weren't being paused when M28 was stalling energy
• Various fixes to Fort Clarke Assault to resolve some error messages and mean that the ahwassas at the start dont bomb the enemy if M28Allies is enabled

Acknowledgements

• Radde - 2 Xander replays
• Hypoxic - mentioning there was an issue with M28 allied logic on Fort Clarke Assault

v313 Update
21 changes, including:

• Ythothas should be more likely to move towards the enemy base instead of attack-moving
• Air slots should be more likely to assist mex upgrades over land factory HQ upgrades
• Upgraded ACUs should be less likely to run from enemy PD
• MAA should be more likely to be built for nearby experimentals, and should then shadow the experimental
• Asfs should be less likely to take a fight they expect to lose to try and save a friendly experimental

Acknowledgements

• Radde - Xander replay which prompted most of the changes this update

v312 Update
20 changes, including:

• Fixing several issues with how M28 plays in the steam version, such as a major bug with the lobby settings mod (that caused AIx to have weird and changing modifiers), and a bug that meant it wouldnt drop plateaus
• More battleships should be built if the enemy has heavy shielding within their range
• Combat units such as ilshies that outrange enemy units should be more likely to kite

Acknowledgements

• Zwaffel - Highlighting a replay where M28 broke due to the Unit Cap Removed mod being used. M28 will now warn at the start of the game that the mod breaks it.

v311 Update

16 changes for various misc issues spotted in a couple of AI vs AI games Unfortunately I lost most of these changes when clicking on an old M28 replay (as I'd forgotten to save them, and loading the old replay caused the M28 code to be wiped), so hopefully I didn't introduce any errors when re-adding them 'blind' (i.e. without being able to test on the replay in question to confirm they worked without errors). Changes include things like:

• Fixing a weird bug that could cause asfs to chase an enemy asf that it has no chance of catching (weird in that the code to stop this had no apparent errors, but would only run if moved to a standalone function)
• making some improvements to strats so they're more likely to ignore low level AA threat and ignore asfs that arent trying to chase them down
• Small improvements if a paragon is built (plateaus no longer have a chance of building a land experimental that cant reach the enemy; combat units are given to teammates to free up unit cap; more engineers should be built and used to build experimentals outside the main base)

Also added one change by request - fatboys (and other land experimentals) owned by M28Easy should, after a delay of a few second, move away from a ythotha lightning storm (so this remains much weaker than M28's normal logic, and also worse than what even a below average player should manage by way of micro)

@c0rtex Make sure you have "UI mod tools" enabled as well as "4z0t's ScoreBoard"

@Nuggets said:

@maudlin27 If a game ender (or even t3 arty, but mostly enders) isnt seen in a game because the game is too short, its kind of an irrelevant comment, no? No offense.

It's very relevant. If a game-ender doesn't feature in games that get to the T4 stage, then the T4 stage on such maps was self-evidently not "too short". It'd potentially also be an indication that the T4 phase could be too long on such maps. It makes no sense to focus solely on the tiny % of games where game-enders feature and decide only based on those games whether the T4 phase is too long. Game-enders serve a valuable role in preventing very long protracted games, and so account should be taken not just of the games they did end, but the games they didn't (that reached T4), and whether that range of time is considered desirable (while also factoring in that even if it was decided that games should be able to go on for many hours like in LOUD, nerfing them into the ground would also decrease the variety of units and strategies, which I'd see as a negative).

The strategy to "rush" enders will still be there, just take longer. Thats the entire point. Give invasions / pushes more of a chance.
Saying pushes / invasions are generally stronger is just objectively wrong. This is not my opinion this is a fact.

It's very much an opinion whether T4 pushes are stronger than game-enders. Something that is fact would be saying something like 'In 4v4 tournament games featuring a large cash prize and top rated FAF players, there were several games where one team focused on a game-ender, their opposing team focused on a T4 push, and the team focusing on the T4 push won' (while an opinion would then be if I said I think that shows that T4 pushes can be more effective than game-enders).

Not for the games I play in (mostly 3v3 mapgens)

• I can't remember the last 3v3 TMM game I had which featured a game-ender, possibly it never has. So no, I don't think it's too short due to them.

• T3 arti gets seen occasionally, but the game is usually over by then (as if one team can build a t3 arti without losing ground significantly then they were likely already well ahead), so I don't think it ends the game too early (and 1 T3 arti can be defended against anyway if the game is close)

v310 Update

• 4 changes from AI vs AI games, including completely reworking gunship retreat logic so they should better avoid enemy ground AA between them and where they want to retreat to
• 5 changes from Radde Xander and fields of isis replays, including fixing an issue that would see M28 send all of its engineers to build a shield instead of assisting an SMD vs an enemy nuke, and lowering requirements for a pre-emptive SMD to be built
• 3 changes from Radde mapgen replays, including fixing an issue that made M28 see too many water zones as its core base (leading to it retreating units to locations that the enemy controlled)

Acknowledgements

• Radde - Various replays on Xander, mapgen, and one on fields of Isis

Trophy award

• Radde and Yew - for beating 1.4 resource M28 on a naval mapgen (the same as the previous trophy)

If I’ve understood you right only shield bubbles overlapping the aoe of a shot take damage? Which sounds similar to steam/non-faf implementation which would make stacking shields much more powerful and hence make turtling more powerful.

v309 Update
24 changes, mostly relating to some AI vs AI games run as part of the 10km version of the 'Vs M28AI' AI tournament, including:

• More shield boats are built if enemy has a novax nearby
• Increased ACU aggression in assassination modes if it's behind on eco
• Improved M28's ability to rebuild power outside of the main base if almost all its core base power is killed
• T3 land should be obtained quicker if M28 has lots of t2 support factories
• M28 should be less likely to upgrade multiple T2 mexes to T3 if it lacks any T3 mexes
• Added a couple of basic chat responses

Acknowledgements

• Nomander - Helping with identifying a function for reading chat messages sent by players (which I’d mistakenly dismissed as only being relevant for resource sharing messages).
• Radde – 3 replays against M28
• Relent0r – 2 replays of team games vs M28