[e54-bot]

Summary

Player events now reach a rule two ways: the scenario events entries the loader already injects, and modeled causes — Damage and Heal raise the player events their effect implies, delivered in occurrence order between rule-instance visits on the same tick. Implements #47 (part of #45's Bastion run-to-completion goal).

• Damage on a living target raises Player Took Damage then, for a player dealer, Player Dealt Damage; a hit that brings health to zero writes the isDead cell and raises Player Died then Player Dealt Final Blow — occurrence order, so Player Died runs before Player Dealt Final Blow regardless of rule file order.
• Heal raises Player Received Healing then Player Dealt Healing for a player healer. A Null dealer binds Attacker to Null and raises no dealer-side event; a Damage/Heal on an already-dead target raises nothing.
• Raised events bind Event Player, Attacker, Victim, and Event Damage/Event Healing (the applied amount). attacker/victim join the event-value table; scenario values supply them as declared player ids (a new loader check) or null. eventAbility/eventDirection remain not_modeled.
• Event team/player filters apply to raised and injected events alike. A busy or Wait-suspended instance queues the event like an injected one; raised events bound at 1024 per tick (a rule whose own actions re-raise its kind would otherwise fan out without bound), past which the run stops not_modeled.
• Event dispatch/eligibility moved to a new events.rs module (size-guard); docs updated (simulator-interface.md, scenario-format.md, architecture.md, findings.md). Cargo.lock syncs the 0.11.1 release version the lock was missing.

Verification

• cargo fmt --check — clean.
• cargo clippy --all-targets -- -D warnings — clean.
• cargo test — all suites pass (new tests/player_events.rs: injected playerLeft runs its rule once with Event Player; killing Damage runs Player Died then Player Dealt Final Blow in occurrence order with rules declared out of order; bindings for Event Player/Attacker/Victim/Event Damage; team-filter rejection; Heal raises received-then-dealt).
• cargo check --locked --target wasm32-unknown-unknown -p athena-sim — clean.
• Bastion acceptance (291 rules): runs to tick 91 and stops at not_modeled: "action wait under a 90% simulation rate" (the documented athena#120 match-clock gap — the next stop, past every player-event construct). No Damage/Heal executes before that stop, so no player-event rules fire in the reference run; with an injected playerLeft on p2 the run additionally executes Bastion's Update targetPlayer when player left rule and reaches the same stop.

Unverified

• No client capture backs the client's event ordering, per-event value sets, or queue bound — all marked modeled, not measured in architecture.md/findings.md.
• The client binding set per event kind is not cataloged, so each event of one cause carries the same Attacker/Victim/amount bindings (e.g. Event Damage reads the killing blow inside Player Died); a probe would be needed to measure the real per-event sets.
• Whether a Wait-suspended Bastion rule ever queues a derived event end-to-end is unexercised before the tick-91 stop.

[e54-bot]

All four review findings are addressed in 0ce5992 (fix(sim): gate Damage/Heal on isDead and correct raised-event reads), pushed to feat/player-events.

1. Damage/Heal liveness gate — fixed. Both actions now gate on the isDead read (the model's death marker, as Is Alive/Set Health read it), falling back to the health pool only where isDead cannot answer. A Killed target takes no pool change and raises no events — no second Player Died, no isDead rewrite; a Resurrected target takes damage and healing events again.
2. Unbound event values on raised events — fixed. EventCtx carries a raised flag; an EVENT_VALUES read a raised event does not bind now stops not_modeled naming the construct (value eventWasCriticalHit), matching eventAbility/eventDirection. Injected events keep missing_input.
3. attacker/victim value types — fixed. The loader accepts a scenario player id string or null and refuses any other JSON kind with a load error, per docs/scenario-format.md.
4. Stale "injected event" comments — fixed in load.rs, scenario.rs, eval.rs, sched.rs, timeline.rs and lib.rs; scenario-format.md, trace-format.md and architecture.md record the contract clarifications (raised-event values are not_modeled, the eventPlayer trace field, the isDead gate).

Verification: cargo fmt --check, cargo clippy --all-targets -- -D warnings, cargo test (all suites pass, incl. new regression tests a_dead_target_takes_no_pool_change_or_events_and_a_revived_one_does, an_event_value_a_raised_event_does_not_bind_is_not_modeled, and the loader-check cases), and cargo check --locked --target wasm32-unknown-unknown -p athena-sim — all clean.

The PR is ready for follow-up review.

[e54-bot]

Follow-up review findings addressed in 3199498:

• Finding 1 (fixed): Respawn/Resurrect now refill the resolved world's health pool to health.max — Respawn's "with full health", Resurrect the same under modeled-not-measured — so a Damage-killed and revived player takes damage and raises Player Died/Player Dealt Final Blow again. The pool move records a world change as Damage/Heal do. New regression test a_damage_killed_then_revived_player_dies_again covers the Damage-kill→Resurrect/Respawn→kill path (seq = [1,3,1,3,1,3]), alongside the existing Kill→Resurrect test.
• Finding 2 (fixed): docs/scenario-format.md loader-check bullet now states attacker/victim event values are a player id or null, matching the loader.
• Finding 3 (fixed): dropped the dead() docstring's last sentence — construct names nothing observable, since world_state carries the stored resolution error and the read's missing_input is swallowed.

Verified: cargo fmt --check, cargo clippy --all-targets -- -D warnings, cargo test (all suites), cargo check --locked --target wasm32-unknown-unknown -p athena-sim — all clean. docs/findings.md records the revive refill in the state-cells bullet.

Ready for follow-up review.

[e54-bot]

Independent review took three rounds. Round 3 found a regression introduced by the round-2 fix: Resurrect now fully heals living targets, which suppresses Player Died and Player Dealt Final Blow on a hit that must kill. Per the orchestration rule, a third round of patches on a design that failed review repeatedly does not converge, so this PR is closed. Issue #47 will be redone from a fresh branch with the findings from all three review rounds as input.