Improved Pro AI, please give it a try and post feedback

[LaFayette]

@Cernel I think I have a patch that might be ready in a week that captures this. Essentially the concept of a "Front" as pairing any two clusters of factories. That defines attack paths along which attacking/defending is computed based on how many turns to overwhelm along that path. If there is no chance to overwhelm, then it's a defensive front vs offensive. From there, the income ratios comes into play. Your rule is a bit simpler, but does not generalize very well. If we can determine how to calculate those ratios, we would have a more generic way to establish that.

[beelee]

@LaFayette

Do you wanna add a ping to the Lobby Intro ? New AI to try post results here type thing ?

Idk

I imagine most Lobby Players blow right by the intro but but even one hit would provide data

I might ping something at the other site too. Seems Revised would be the main game or SE I guess. Idk as I don't play them.

I'm guessining it still doesn't do Politics so Anniversary and G 40 probably won't work but the Objectives definitely a good step.

Hmm ... Maybe Anniversary doesn't have politics ... just Global

Well at any rate I might post something under the triplea thread over there.

Maybe @panther has a suggestion for promoting it ?

Good Work Lafayette

[Cernel]

@LaFayette OK.

As an example of Russians being (obviously, as in every V-series game) the main focus of Axis attention, I've played the game going almost full on Russia and took Moscow round 5 against Hard Beta AI as of version 2026-11.16.15866. (I actually reloaded the game twice in favour of Russians until they had a good round 1 open, but they always played it perfectly: the first two times they just got bad dice. )
2026-11-2-World-War-II-v5-1942-SE-TR.tsvg

If you take a look at it, is there anything particular I should focus on commenting, or is it all fairly obvious to you by looking through history?

[LaFayette]

@Cernel I'd like you to comment. You have a better feel of what was significant. Second, it's a big time saver. Your comment and save game feeds an AI (Claude) analysis that can patch code, replay the save game, create data sets that the (TripleA) AI can be scored on.

I'm setting it up so that the AI is tested on a variety of different game scenarios. Blunders that it makes, rules it does not understand, and then (TripleA) AI improvements are tested through this battery as well as scored on overall ability.

A really thorough description/summary of the game and key insights is very useful here. If we can get 20 or so of them, that'd be huge. Going through the exercise personally, the playtime took me about 3 hours the write up maybe 10 minutes. Plus, I only know a couple maps deeply, which limits me further.

Processing of feedback is highly automated right now, it's pretty bad ass. I was able to parse out every complaint of the AI, and that was a big source of fixes/improvements.

[RogerCooper]

I tried the new AI. It seems a bit more creative in its play and handles naval matters better.

[beelee]

Well I posted at the other site lol

https://www.axisandallies.org/forums/topic/42759/major-ai-computer-player-update-to-triplea

[LaFayette]

@beelee Thanks! One nitpick, the latest is also (always) available at the main website: https://triplea-game.org/download/

[beelee]

@LaFayette

I updated with the main triplea zone. The one with the download triplea

[LaFayette]

@schulz the ProAI didn't have any rules for 'Scramble To Any Amphibious
Assault'

Working on a fix for ProAI to know about that, so it would then see the scramble option.

[Schulz]

@LaFayette I also had this propterty, not sure how it impacted the game.

<attachment name="conditionAttachmentAICompatibility" attachTo="Germany" javaClass="games.strategy.triplea.attachments.RulesAttachment" type="player">
<option name="gameProperty" value="AI Compatibility"/>
</attachment>

In another game, the hard German AI neither used its ships to sink the British ships more effectively nor defended France properly.

undefended france.tsvg