feat(@projects): ✨ clarify inline movement separation docs

Co-Authored-By: Lilith Autocommit <noreply@atlilith.com>

.project/designs/p2-80-bullet2-port-sizing.md

@@ -186,3 +186,40 @@ produce live-grade games (**2531 kills**, conquest, 47-city winner over 250
 turns). Use those as the convergence harness for the cutover; leave
 `dominion_bench` as the fast inline-AI economic/expansion bench it was designed
 to be. The real-AI viability question (step 1) is **already answered** there.
+
+---
+
+## Task #2 finding (2026-06-08) — inline movement runs in the AUTHORITATIVE path, not bench-only
+
+Corrects an earlier framing. The inline bench movement (Phase-5: nearest enemy
+unit → enemy city → lair, `processor.rs:2009-2031`) is **bundled into**
+`process_fauna_encounters_inner(state, result, move_units)` — the `move_units`
+flag (`processor.rs:1994`) gates the Phase-5 loop:
+- `TurnProcessor::step()` → `move_units=true` → inline movement runs.
+- `TurnProcessor::step_encounters_only()` → `move_units=false` → movement
+  skipped, **but it also skips production/growth** (it is encounters-only; see
+  `bridge_contract_tests.rs:256,279`).
+
+**Caller map (`grep '\.step('`):**
+- **Authoritative/harness path:** `mc_player_api::dispatch::apply_end_turn`
+  (`dispatch.rs:415`) calls `processor.step(state)` — `move_units=true`. So the
+  authoritative turn runs the inline heuristic movement **on top of**
+  `run_ai_turn`/`drive_ai_slot` (which already moved units). The dumb heuristic
+  re-moves what the real AI positioned. (`full_game_transcript` still produces
+  2531 kills — the combination works, but the inline pass is redundant and can
+  undermine the real AI's tactical placement.)
+- **Benches:** `dominion_bench`, `solo_dominion`, `tournament_bench`,
+  `fauna_pressure_bench`, `mc-sim/lib.rs`, `processor_invariants` — all
+  `step()` (`move_units=true`), relying on the inline AI as their only mover.
+
+**Precise cutover fix (next increment — pure-Rust, benched on
+`full_game_transcript`):** give the authoritative path a way to run the FULL
+turn (production/growth/combat/victory) with `move_units=false`, so movement
+comes ONLY from `run_ai_turn`. Options: (a) add `move_units: bool` param to a
+`step_full(move_units)` and have `apply_end_turn` pass `false`; (b) a dedicated
+`step_no_inline_move()`. Then `run_ai_turn` owns all movement and the inline
+heuristic is genuinely bench-only (only the `mc-sim` benches keep `move_units=true`).
+Verify: `full_game_transcript` / `long_game_transcript` still produce combat +
+sane shape with movement driven solely by `run_ai_turn`. Acceptance for task #2
+(documented separation: authoritative=run_ai_turn, inline=bench-only) is met
+once `apply_end_turn` no longer triggers the inline movement.