# Baseball implementation review 2026-11-02. Independent read-only review of BaseballGadget, SceneBaseballs, OriginalBodyPhysics, LabSimulation, PlayerSceneBinding and Program wiring. Source baseline/evidence and provider limitations are in [baseball.md](baseball.md); no unchanged REA queries were repeated. ## Resolved findings 1. **Companion cleanup policy differs by gadget.** The initial implementation subscribed Baseball.EquippedChanged to InvisibleEquipped, incorrectly skipping restoration in LEV6/LEV7/LV1C/LV1D/LV4A/LV6A. Source raw6 cleanup42871D..878D restores Goddard when its pointer exists without Invisible's9E8 skip test (compare428144). PlayerSceneBinding now uses a dedicated BaseballEquipped handler with unconditional restoration. A regression verifies all six levels. 2. **Initially hidden tutorial trophy contact must be disabled.** At review time SceneBaseballs activated its parent with Active=Visible=true, while SceneInteractions3TRO contacts used Active regardless of visibility or pair enable. This permits collecting VR07's initially hidden trophy before earning it. Primary subsequently added a narrow trophy pair-enable/activation correction in SceneBaseballs and requires trophy pair enable in SceneInteractions. The regression exercises real ball/target contacts, strict fourth-hit activation, and HasCollision0/−1 behavior. It passes. Do not replace this with a scene Trigger: target430570 calls parent428(1). The source alias needed for the second correction is proven: pickup property reader40D41A passes interface+584 and literal4ED26C=`HasCollision` to3FC, type6. C3DVRTrophy interface table4BE404 has428→40E1F0. For argument1 it calls 3AC→4730E0→224(true) only when HasCollision is nonzero, then unhides through complete58(0) when needed. Argument0 calls3A4→473030→224(false), then hides. It does not enable214 integration, award score or invoke the script interpreter. These are supplemental reviewed instructions; primary owns adding the exact guards to baseball_targets.py. They are not attributed to new REA pseudocode. ## Checked source contracts - Equip has no fresh energy gate. Camera height adds30; cleanup restores200 through170→437740 and clears pitch to0. Ordinary non-raw6 input subsequently restores pitch10. Current Forward/Backward mapping correctly increases pitch on Up and decreases on Down,20*fixed-dt; this source keyboard field is unrelated to PageUp/PageDown. - First supported hold shows the aim strip and arms without refreshing saved pitch. Later holds clamp0..45. DisplayPitch is copied before capture, matching the original one-invocation label delay. Release hides the strip and launches once from the saved pitch, including after unequip and at zero energy. - Cleanup retains pending throw, saved pitch and20 projectile entries. Source ring reuse does not clear pending impulse/support; current Launch preserves both. Modern scene reset clears the pool and is an explicit lifecycle boundary. - Projectile launch uses local(0,80,50), normalized local aim slope pitch/45, source quick magnitude and speed1800. Upright yaw cases pass. Continuous full-Euler transform remains a declared original table-precision boundary. - Ball floor friction and wall damping both read slot248/live150=.46; authored24=3.2 belongs to a different field. Floor support uses radius15, factor1, prevent-lowering=false. Current Projectile.Step explicitly supplies.46. - Landing queues oldX,−.65*oldY,oldZ into impulse before support zeros velocityY; next body consumes it. Current implementation preserves this phase order. - Program runs ball/target bodies before player input, and target frame callbacks afterward. Input ownership follows LabSimulation's controlled path. Release from a pending latch is processed even when raw6 is no longer equipped. ## Declared limits, not release claims SceneBaseballs uses per-tick sphere overlap instead of original pair membership, registry order and swept discovery. It can re-hit an overlapping ball after a target respawns, whereas original enter callback timing still needs the pair scheduler. Whole-scene pair response and other ball consumers remain incomplete. Ball angular wall response is omitted, and each managed ball tick supplies zero angular state; its billboard's positional path is the bounded tested subset. Only Canv2 is used because no four-canvas animation advancement was recovered. The full original game and native Windows input have not been runtime-compared by this review. ## Verification `dotnet run --project tests/GamePort.BaseballTests --no-restore` passed11 groups with the local .NET environment. Python discovery `test_baseball*.py` passed16 groups covering source guards, resources and target projections. These results were rerun after the companion/trophy fixes. No remaining concrete blocker was found within this bounded review; the declared limits above remain. No runtime or shared source was edited by this reviewer.