Fix AI target selection for DamageSource::Target to prefer lethal rec…

[CodeOptimist]

Intro

This intro written by the human (me). Note that the Model/Tier may not be in the Frontier table. My expectation is Matt will override the check.

Also note that we actually did follow the /engine-implementer pipeline for quality, even though it "wasn't necessary". I can gladly provide the reports/output from each of those steps. This may not be a good place to attach them though.

Edit: Maybe I should have changed the implementation method line to say so, with a note that it's technically N/A? 🤷
Edit 2: Occurs to me that I could have moved the Model/Tier to this intro... since that's a guard against fully autonomous PRs? 🤷 But I also wouldn't want this automatically merged, since usage of DeepSeek is under evaluation, so maybe as-is was best.

Summary

Fixes a bug where the AI selects non-lethal targets for self-damage spells like Self-Destruct. It updates the AI policy in removal_lethality.rs to correctly resolve the bound damage source for DamageSource::Target effects, enabling accurate lethality scoring.

Files changed

• crates/phase-ai/src/policies/removal_lethality.rs
• crates/phase-ai/src/policies/tests/removal_lethality.rs
• crates/phase-ai/src/search.rs

Track

Developer

LLM

Model: deepseek-v4-flash-0731
Tier: Frontier
Thinking: high

Implementation method (required)

Method: /engine-implementer
(But technically not-applicable — Changes are strictly AI-policy scoring updates entirely within crates/phase-ai/, touching no crates/engine/ game logic.

CR references

CR 120.1, CR 120.3, CR 120.3d, CR 120.3e, CR 208.1, CR 601.2c, CR 608.2c, CR 702.2b, CR 704.5g, CR 704.5h.

Verification

• Required checks ran clean, or the exact CI-owned alternative is stated below.

• Gate A output below is for the current committed head.

• Final review-impl below is clean for the current committed head.

• Both anchors cite existing analogous code at the same seam.

• cargo test -p phase-ai --lib self_destruct_target_selection_prefers_lethal_over_nonlethal_body — ok

• cargo test -p phase-ai --lib removal_lethality — 25 passed

• cargo test -p phase-engine self_destruct_target_power — 3 passed

• cargo clippy --all-targets -- -D warnings — clean

Gate A

Gate A PASS head=d4ddd2042cf92642e1c5b4fb181a224b8985d226 base=c44a4512e6f91684068c259a028eb44b4d801340

Anchored on

• crates/engine/src/game/effects/deal_damage.rs:1671 — analogous source resolution authority (source = first object target)
• crates/engine/src/game/quantity.rs:4940 — analogous amount resolution authority (ObjectScope::Target resolves first object target's power)

Final review-impl

Final review-impl PASS head=d4ddd2042cf92642e1c5b4fb181a224b8985d226

Claimed parse impact

None.

Scope Expansion

None.

Validation Failures

None.

CI Failures

None.

Summary by CodeRabbit

• Bug Fixes

  • Improved AI lethality evaluation for effects that deal damage based on the selected source’s power.
  • AI now correctly evaluates target choices after source targets are selected while remaining neutral when the source is unresolved.
  • Fixed target selection for Self-Destruct-style effects, allowing the AI to choose lethal targets and avoid nonlethal ones.

• Tests

  • Added coverage for lethal, nonlethal, and unresolved source-target scenarios.

[coderabbitai]

📝 Walkthrough

Walkthrough

DamageSource::Target now resolves from bound target selections. Power-based target damage uses the selected source’s power for lethality evaluation. Unit and end-to-end tests cover lethal, nonlethal, and unresolved-source cases.

Changes

Target-sourced lethality

Layer / File(s)	Summary
Bound source resolution crates/phase-ai/src/policies/removal_lethality.rs	Resolves DamageSource::Target from the first selected target when available.
Power-based damage evaluation crates/phase-ai/src/policies/removal_lethality.rs	Resolves target-sourced quantities against the bound source and target slice. Other damage sources retain existing behavior.
Self-Destruct regression coverage crates/phase-ai/src/policies/tests/removal_lethality.rs, crates/phase-ai/src/search.rs	Adds configurable source power, bound-source fixtures, lethality cases, unresolved-source coverage, and an end-to-end target-selection test.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Sequence Diagram(s)

sequenceDiagram
  participant AI
  participant TargetSelectionProgress
  participant RemovalLethality
  participant GameState
  AI->>TargetSelectionProgress: select source and recipient
  TargetSelectionProgress->>RemovalLethality: provide bound selection
  RemovalLethality->>GameState: resolve source power
  GameState-->>RemovalLethality: return damage amount
  RemovalLethality-->>AI: score target lethality

Loading

Possibly related PRs

• phase-rs/phase#6665: Extends removal_lethality with DamageSource::Target handling and power-based source resolution.
• phase-rs/phase#6826: Modifies AI lethality evaluation for target-sourced, power-based damage.
• phase-rs/phase#6852: Adds selected-target resolution and Self-Destruct lethality tests.

Suggested labels: bug

Suggested reviewers: matthewevans

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly describes the main change: improving AI target selection for DamageSource::Target effects to prefer lethal targets.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[matthewevans]

Closed without implementation-diff review.

This PR was opened on 2026-08-07, after the 2026-07-24 Frontier-tier cutoff. Its canonical PR-body declaration is Model: deepseek-v4-flash-0731. The policy version already present in this PR's base at creation requires Frontier-tier models only and lists the accepted vendors/models; DeepSeek is not an accepted Frontier model. Tier: Frontier does not override the required accurate Model: declaration.

This is a policy-routing decision, not a judgment on the code. Please rerun the work on an accepted Frontier-tier model and open a fresh PR from current main with an accurate canonical model declaration.

[CodeOptimist]

Do I ping you @matthewevans (human) or do you get notified of this on your own? 😆

[matthewevans]

Maintainer note: PR is approved for getting a formal review. Do not close.

[matthewevans]

Approved for the current head d4ddd2042cf92642e1c5b4fb181a224b8985d226.

Review evidence: the independent implementation review found no findings; CodeRabbit has no actionable feedback; and current Rust, frontend, card-data, WASM, paired-seed AI, and decision-cost performance checks are green. The fix is a scoped AI lethality-policy correction with discriminating lethal, nonlethal, and unresolved-source coverage.