# Jimmy landing → idle orientation ownership 2026-11-05. User reports a slight bank/pitch can remain after landing without further movement. This investigation tests the original paths rather than assuming that support implies an upright snap. Parent owns runtime integration. ## Evidence and coordinate owners Original Neutron.exe SHA256 `f5281076e1a2492f2d61f87aae3f4544b6d0f535ebe33f9fbb6f78710fe1aea8`. Retained complete REA instruction Evidence `ev_0d64cc58fa8f5573fc8e61e0f2c20e36288099212bd2257d3a3121361fd6c904`, artifacts/jimmy-movement-audit/rea-original-disassembly.json. This report reuses that observation; it does not claim a new Ghidra decompile. VA-to-file mapping below is checked through PE sections (these EXE text offsets equal VA−400000). JN-Engine's original C3DJimmy/C3DPlayer decompiler notes supplied leads only. Current Euler is game224/228/22C (complete2E4/2E8/2EC). Destination Euler is game274/278/27C (complete334/338/33C); game+A (completeCA) is facing dirty. Destination setter36C→472660 storesXYZ and sets dirty. Jimmy facing43A420 reads current/destination independently and runs before controlled-player gating. There is no small-error snap in that recovered interpolator. ## Grounded and idle transitions | Original VA / file offset | Recovered behavior | |---|---| |438A60..438B9E /38A60..38B9E|Input dispatch uses retained movement mode game704. Mode0 selects ordinary ground handler438BC0; mode1/2 selects439900. It does not itself reset current/destination Euler.| |438BC0..43915D /38BC0..3915D|Complete ground handler prefix samples input, support/root, idle animation/sounds and camera-vector angles. No330/32C/36C Euler dispatch or direct current/destination Euler store occurs in this bounded prefix.| |438F66 /38F66|Ground accepted-heading flag ESI starts0. Keyboard/accepted analog directions set it; neutral input need not do so.| |43951E..439520 /3951E..39520|If ESI remains0, jump4395CF, skipping all destination writes. This preserves old destination pitch/roll/yaw and dirty state.| |43957C,439599,4395C9 /3957C,39599,395C9|Accepted steering publishes pitch0/roll0 and the selected destination yaw through36C. Input, rather than the mere fact of standing, owns these upright destinations.| |437AEA,437B12 /37AEA,37B12|Grounded/released wrapper calls complete140 when prior movement mode1, before the animation decision and resetting mode0/turn0 at437B9E..437BAC.| |429910..429962 /29910..29962|Complete140 clears complete1E64/1E68, hides the A18 menu indicator except rawmode2 and stops A3C sound if valid. No current/destination Euler write occurs. The menu488 target403700 only hides two UI resources.| |43B28A..43B2EB /3B28A..3B2EB|Accepted STOP selection uses generic animationE0 and clears acceleration278. It does not reset current/destination Euler; same-animation request returns earlier43AFF8..43B010.| |40DD90..40DF79 /DD90..DF79|Generic animation start resolves the named morph, starts it and sets shape/timing. No Euler destination/current setter is dispatched in the resolved routine.| |game37C→472970 /72970|Jimmy's body landing callback binding is the bare no-op/log stub, not an upright reset.| The derived Jimmy input hook426030 also has a separate impact latch. Its support query is4260AA, falling gate uses vertical thresholds at4260EF/4261B1 and sets **game1D98** at4261C5 (complete1E58). On support42622D..42625E, it clears that latch, plays soundC0 and calls complete1A0(-10,-10,5), the camera effect path. This is not a rotation reset. Offset names in the friend's mixed-interface notes must not be copied without the C0 adjustment. The pickup-halo AddEuler4264F8 writes the effect object1D94, not Jimmy. The426F47 Euler copy likewise targets an attached action objectA10 from Jimmy's current Euler. Ordinary air input439900 independently accepts support for speed decay, but turn return439989..439A58 has no supported-axis reset. Its ordinary destination 439DB7..439E2D is `(acceptedPitch,currentYaw,-storedTurn)`. The last accepted air input can therefore leave a small bank destination after turn0 is subsequently written by the grounded wrapper. Neutral ground input alone does not replace it. Earlier [Bubble recovery](bubble-motion.md) already records the analogous no-input retention; this is not evidence for introducing a cleanup timer or snap. ## Rendering and scope 473540 converts current degrees to source14-bit angle units. The existing JimmyRenderTransform already performs that quantization. The OMT hierarchy retains full pitch/roll. The final OMT model-matrix decomposition remains its previous precision gate; none of these static paths establishes a general "grounded means upright" policy. **Recovered:** the local writes, dispatch order and retention contracts above. **Inference:** stale last-air bank can explain the reported freeze in the port and is also possible through these source components. A full original frame trace is still required to prove whether another scene writer intervenes in the reported route. A component result is not that whole-game observation. **Implementation:** parent owns any runtime change; do not silently insert a landing snap and label it original behavior. The independently recovered CAM/MCA flattening writes both current and destination and has a different owner. ## Guard and finite component verification `tools/binary/jimmy_landing_rotation.py` guards original SHA, seven branch/field anchors, five virtual bindings and eleven bounded range hashes. Optional retained REA input verifies Evidence identity and lists all ground-prefix calls. `jimmy_landing_rotation_probe.c` executes the original43915D..4395CF keyboard suffix and actual43A420/458F90/472480 facing leaves with real472660 destination setter and4723F0 getter. A synthetic actor/input phase replaces body, camera, animation resources and full scheduling. It seeds a supported neutral-input boundary after the last airborne destination publication, then records one and 300 input/facing passes; it does not execute full landing or whole gameplay. All seven finite original-component cases complete successfully inside the network/filesystem-isolated Wine32 generated copy. The final capture records wineExit0 and `JIMMY_LANDING_ROTATION_COMPLETE`. Example observations: - Neutral last-air target `(0,155,-5)` retains that destination; after300 passes, current is approximately `(0,155,355)`. The five-degree bank survives. - Neutral last-air forward target `(-40,290,-25)` retains it and settles to `(320.000183,290,335.000183)`, rather than upright. - Dirty0 with current `(10,155,15)` leaves that rotation unchanged for300 passes. - Active Up publishes `(0,290,0)` and levels via the original interpolator; Left likewise publishes zero destination pitch/roll while changing yaw. The neutral suffix does not query support itself: caller selection of the ground input path is the fixture boundary, not a newly observed whole landing callback. `tools/binary/jimmy_landing_rotation_reference.json` retains all7 observations, source range/byte/virtual guards, generated C/executable/shared harness/log hashes, and capture exit metadata. Raw logs, generated source and executable are retained under artifacts/jimmy-landing-rotation/. NoOriginal directory is mounted for the probe and no original entrypoint executes. This confirms retention in the specific original components and leaves another whole-frame writer explicitly unresolved; it supplies no evidence-based justification for a universal land snap. Reproduce static preparation: ``` python3 tools/binary/jimmy_landing_rotation.py ../Original \ --generate-probe artifacts/jimmy-landing-rotation \ --rea-disassembly artifacts/jimmy-movement-audit/rea-original-disassembly.json ``` Compile generated C with local i686-w64-mingw32-gcc and `-Itools/binary`; execute only through tools/recon/reference_capture.py's `--asset-probe` boundary with `--probe-success JIMMY_LANDING_ROTATION_COMPLETE`. Parent owns regression-test placement and runtime decisions; no shared/runtime file was edited by this agent.