Bot war features

[KonH]

let's discuss about bot war usage
bot should declare wars and tried to achieve profit from it (change control and population -> score for itself and enemies) using new war-related cards, ask all clarification questions and decompose changes carefully first

depends on:

• War mechanics (core) #69
• Declare war card #70
• Sell arms card #72
• Ultimatum/Surrender cards #73
• Peace resolution: provinces change owner, control changed, gold received #74
• Revenge card target to specific country #76

[KonH]

Automation hit a session/usage limit on claude. Salvage: clean (working tree clean). Rerouted to cursor immediately instead of waiting until 2026-08-01T11:40:00+00:00.

[KonH]

Handoff — Bot war features (#83)

Stopped at the specify approval gate (no plan/code). Discussion + decomposition only, per the issue prompt.

Artifacts

• Spec: Docs/Specs/26_08_01_09_bot-war-features/spec.md
• Branch: feature/bot-war-features
• PR: Bot war features #110

Decomposition (provisional)

1. Phase 1 — Observation extensions (wars, rivals, progress, TargetCountryId, score signals) — outside /implement-bot-feature; needs /specify+/plan
2. Phase 2 — Declare / revenge profit-gated play of declare_war / revenge
3. Phase 3 — Prosecute / resolve sell_arms / ultimatum / surrender
4. Phase 4 — Arbitration with discoverAndControl (order, gold reserves)
5. Phase 5 — Eval under Docs/BotFeatures/<featureId>/

Key finding: IBotObservation has no war/relation/progress/score fields today, so a pure IsPlayable-heuristic feature cannot implement score-aware war strategy without Phase 1.

Remaining work

• Owner answers the Ambiguities in the spec (22 items)
• Approve/revise the phased decomposition and issue boundary
• Then /plan (and only after that, implement)

Clarification questions (please answer)

1. Packaging: one warProfit feature vs split declareWar + prosecuteWar (+ revenge)? Profile order vs discoverAndControl?
2. Profit objective: maximize own org_score, own−rival relative, gold mix? Which rivals?
3. Declare threshold: when is war “worth it” (control floor, CountryScore skew, expected Δscore vs gold)?
4. Unlocking: also play make_rival / opinion cards, or only act when declare_war is already playable?
5. Arms vs force: prefer sell_arms first, or force with ultimatum/surrender as soon as score-positive?
6. Intentional lose: confirm surrender only when opponent-win helps the bot’s score; any hard vetoes (HQ / high MyControl)?
7. Revenge: always avenge when eligible, or only if still profitable?
8. vs discoverAndControl: hard priority / soft Δscore threshold / gold partition / separate profiles?
9. Observation depth: full seat-visible war API vs minimal ActionId+TargetCountryId heuristics?
10. Eval settings: longer than ~5y / more seeds / mid-rivalry setups?
11. Bot control floor beyond card gates for declare/prosecute?
12. Relative hurt target: specific rival org vs all others?
13. Gold policy: shared minGoldReserve vs per-action budgets given costs 100/200/300/500/50?
14. Issue boundary: is Phase 1 in Bot war features #83, or a separate prerequisite issue?
15. sell_arms drift: issue text said “+300 gold”; shipped config is cost 200g + damage bonus only — restore grant, or bot against shipped economics?
16. Natural vs forced peace: wait in extreme progress bands, or always force when playable + score-positive?
17. Partial discovery: prosecute with only one side discovered?
18. Occupation EV: expose occupation for transfer estimates in v1, or control×CountryScore only?
19. Plays per day: hard-cap war to one play/day; can discover still act same day?
20. Eval opponents: passive / baselineCardPlay / discoverAndControl / mirror war bots?
21. featureId naming preference?
22. First-ship success bar: “sometimes plays war cards” vs “beats discover-only twin on org_score” vs “wins wars on controlled side”?

Please comment answers (or point at revised defaults in the spec). After that, remove ai-need-attention to resume planning.

[KonH]

answers for now:

1. separated features, discovery/control and war features should be selected by scoring (in some cases war is more important, in some cases it is better to discover and control)
2. org_score
3. better org_score change is most important, prefer wars with high win % (calculate it using new shared helper, will use it later)
4. use make_rival also
5. sell_arms first
6. surrender can be played on opponent country, it is intended use
7. if profitable
8. see p.1
9. see p.3
10. explain
11. depends on card conditions, you should also use improve military advisor and ruler opinion cards to achieve ability to play related cards
12. best score org
13. explain
14. we will split to separated issues after decompose
15. based on configs
16. not sure, let's discuss, but ultimatum and surrender could be played to limit loses in specific country
17. yes
18. explain
19. not per day, per hours configured in config
20. explain
21. decide yourself
22. let's discuss

blocked considering major feature cut planned

[KonH]

feature cut completed, now sync with main and re-check specification considering changes in new commits and clarification questions above

pay attention to:
0) make rival card - it should be used when make sense to create new rival to war with it

1. new card draw flow allow to select 1-of-3 cards, bot need to check what better fits it's intentions
2. card discard is embedded to control feature, it is not right - should be common mechanism to discard when hand don't have suitable cards
3. declare war require military advisor opinion, sell arms increase win chance - bot needs to support that complicated flow
4. countries could be destroyed - it may be good option to reduce enemy scores in addition to reducing enemy org score via enemy controlled country score reduce
5. war should be used carefully considering control and score change if bot have control both sides of war, so potential profit calculation should check that

ask new clarification questions (if any) before plan