Files
Neutron/docs/gameplay/jimmy-air-bridge.md

120 lines
7.4 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Jimmy airborne targets and rendering bridge
2026-11-05. Continuation of [movement stabilization](jimmy-movement-stabilization.md).
Preserve the playable controller baseline and original550 impulse/−800 gravity;
JN-Engine is used only for original-code navigation, verified with REA/source.
## Recovered contracts and integration
[Target state](jimmy-air-target-state.md) records source addresses/offsets, Init,
accepted input, pitch seeds, destination/dirty ownership and route retention.
[Render transform](jimmy-render-transform.md) records source-degree/OMT matrix
recovery and copied-original observations. These distinguish instruction/data
recovery, native component execution, implementation and port verification.
- `JimmyAirMotion.Destination` matches all12 previously recovered ordinary-air
writes: Up reads retained forward pitch, Down reads backward pitch, Up wins,
neutral pitch0, yaw=current and roll=−turn. These are now live.
- `LabSimulation` retains forward/back pitch seeds. Constructor/debug reset seeds
−40/+10; ordinary air apply reasserts−40/+10, Bubble apply−10/+10. Input reads
whatever the preceding apply left. Bubble equip/cleanup does not reseed.
- `PlayerLevelState` carries these two process scalars only through its already
bounded lab/VR04/outside ordinary/jetpack route. Init resets targetEuler0 and
dirty=false independently. Original TSK files gain no fields. Debug reset R
remains a modern constructor-style reset.
- Current Euler, destination Euler and dirty state are separate. Accepted active
ground input writes upright destination; neutral leaves it. Air writes each
pass. Facing runs before raw control gating, only when dirty, then retains dirty
according to the before-step equality check. Authored fullXYZ writes current
Euler without replacing destination. Default route STRTs are upright; arbitrary
selected STRT Euler is now passed in full instead of dropping X/Z.
- `PlayerSceneBinding` and application sync publish full current Euler. The old
Bubble-only orientation switch/seeding bridge is removed. Vehicle passengers
continue to copy their full authored Euler; Scooter's existing copied-destination
adapter is retained as its separate compatibility boundary.
- The character model now uses `JimmyRenderTransform`: source-degree quantization,
proven local rotation order and the unrotated logical root. The former sole
correction becomes a fixed model-local translation before rotation. Upright
placement is preserved; tilt pivots around the logical root. This correction
remains modern, not an original recovered offset or complete pivot parity.
Debug noclip/Stop Motion explicitly freezes the pending facing request along
with body motion, so enabling/debug-stopping flight cannot retain an unseen tilt
request. This is modern tool policy, separate from original script preparation.
Script STOP still preserves destination and queued body/effect work.
## Component and integration verification
12 destination observations compare against the managed target helper. New
integrated cases cover Init/dirty/current writes, control locks, ordinary tilt,
scalar versus horizontal velocity, fresh/retained first Bubble input, cleanup,
route carry/reseed and noclip.28 movement groups,20 body groups and13 Grappler
integration groups pass in targeted runs. Lab and underground input-only routes
pass. Renderer component/whole-regression/native/package results follow below.
The old powered-path regression required50 units of lateral displacement under
flat air. Source pitch attenuation produces about38 units on both paths in its
same60-step fixture; the test now requires30 measurable lateral units and retained
lift. No gameplay tuning was changed to recover the old flat-air distance.
Artifacts: artifacts/jimmy-air-bridge/{movement-diagnostic,movement-targets,
movement-bridge,body-targets,grappler-targets,lab-route,underground-route}.log.
## Remaining gates
The source hierarchical transform is recovered; OMT's later final render-matrix
decomposition/re-quantization is still approximated, with measured seam deviation
recorded in the transform report. The model-local sole alignment and original
animation-root/completion behavior remain distinct. No radius, support, dynamic
or special-contact admission changes are made here. Root-changing callback Vec4
publication, world-response versus integration, hazards/climbs and dynamic pair
eligibility remain the next Jimmy body checkpoint. Whole-original trajectories,
original cadence and native Windows gameplay still require comparison.
### Native movement integration
Native Linux GTX1070/private Xvfb1920×1080 passes300-step VR04 powered steering
and240-step lab walking for shift/right-shift/insert/keypad0/right-mouse, including
actual SDL event receipt and resize. New smoke captures at powered step75/110 and
ordinary jump step90 record current/target Euler and the rendered frame. All25
final/airborne readbacks match across bindings, with zero Vulkan validation errors.
The first forward/brake/ordinary airborne captures were visually inspected.
Powered step75 has currentEuler≈(323.858,293.333,349.214), target(−40,292.917,−25);
step110 has≈(5.430,307.417,342.110), target(10,307.333,−5). Current Euler continues
through its own interpolation/turn phase rather than snapping to the target.
The300-step fixture travels359.19 versus the prior flat-air403.88; maximum rise
418.93, one burst/two landings and VR energy20 are preserved. This is port replay
measurement, not original whole-game trajectory evidence. Native Carl input lock
and control-independent body/fire/resumption also pass. Logs, images and all25
hashes: artifacts/jimmy-air-bridge/window/.
### Whole regression verification
All checks in scripts/test.sh complete across the main invocation and recorded
resumption:325 Python tests,2 HUD extraction tests,44 managed projects,36 main
Vulkan render completions, asset decoder checks, save/load, animation and SDL
dummy audio/lifecycle, plus539 unchanged originals. The initial invocation stopped
at a Director fixture that expected a tilt to survive five active ground-input
steps without full-Euler sync. Its corrected fixture seeds current tilt immediately
before the command, preserving the test's intended camera-vs-AI preparation
boundary; the Director suite passes. The resumed tail is retained separately.
This records the resumed checks, not a successful exit of the initial invocation.
Logs: artifacts/jimmy-air-bridge/full-tests.log, director-corrected.log and
test-tail.log. No gameplay math was tuned to restore obsolete fixture assumptions.
### Packaged checkpoint
Both self-contained packages identify clean runtime commit
`27be8ed63079ae2e3bf7e965ee205c4a57a96559`, source_dirty=no. Each contains519
unchanged original asset copies and4271 scene/resource files matching preparation.
Artifacts: package.log and package-identity.json. Standalone Linux from /tmp with
DOTNET_ROOT/DOTNET_CLI_HOME/NUGET_PACKAGES unset passes right-mouse powered flight
and Shift walking with actual SDL input/resize, zero Vulkan validation errors.
All five final/forward/braking/ordinary-air readbacks match development byte-for-byte;
package-{walk,jetpack}.log and package-readbacks.json retain evidence. Post-package
original audit verifies all539 hashes unchanged. Windows cross-build/resource
checks remain distinct from native Windows and whole-original gameplay verification.
Subsequent documentation commits record validation; package BUILD-INFO preserves
its exact runtime source provenance.