Files
Neutron/docs/gameplay/jimmy-movement-stabilization.md

148 lines
9.1 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 movement stabilization
2026-11-04. User pauses gadget expansion and prioritizes solid Jimmy movement.
Current C# movement is the playable baseline; runtime packages fccc0b9 and source
42ea2b8 remain preserved. JN-Engine's native/Godot movement is explicitly not the
behavior oracle. Its saved original decompiler/class index is a navigation aid;
REA, authoritative bytes and copied-original observations verify corrections.
## Working checkpoints
1. Compare ordinary ground/air input, velocity and facing with copied-original
cases; preserve current acceleration/jump/gravity values unless contradicted.
2. Verify jump/hold/release/landing, jetpack acceleration/expiry, queued impulse
and physics-before-input ownership, including fresh/retained scene order.
3. Recover lab/outside contact metadata, root publication and animation-owned
response sufficiently to replace the provisional capsule safely. Preserve
scene/actor state ownership; do not enable unsupported special contacts.
4. Verify the combined input/physics/control/animation behavior on lab → VR04 →
lab → outside, including all jump aliases, stairs/slopes/walls/ceilings,
cutscenes, reset, RETURN and stable presentation-rate behavior.
## Evidence boundary on entry
Current ordinary body is source-backed in bounded tutorials; lab/outside still
use the modern capsule. Controller input and apply are simplified in LabSimulation,
with source-based Jump550/gravity800 plus jump acceleration correction, fixed60Hz
as a modern policy, and a sole/root render bridge. Previous numeric/route/native
verification remains valuable; it does not certify complete movement fidelity.
Ground/air/math corrections are separate from admitting special/dynamic physics.
New independent reports: jimmy-ground-motion.md, jimmy-air-motion.md,
jimmy-orientation.md and ../checkpoint/jimmy-special-contact.md. Parent owns
integration/jump/route verification; investigators own their named new files only.
Findings/tests and implementation status are updated here as chunks finish.
## Controller checkpoint, 2026-11-05
Integrated `JimmyGroundMotion`, `JimmyAirMotion` and `JimmyOrientation` into
LabSimulation. Ground Up/accepted Down publishes current heading even without
turn input, clearing the old turn destination; neutral retains it. Acceleration
preserves existing overspeed. Ground backward acceptance honors raw−1/3 primary
fire. Air input retains the shared keyboard-turn latch, performs its independent
supported subtraction/brake and compares coast with the wide product. Full-Euler
apply preserves incoming Vy and correctly gates lateral/yaw on incoming velocity.
Facing uses original shortest arcs/full-axis angle normalization and runs in the
base phase even without raw control. Published speed, pending impulse, effect
clock and fresh/retained ordering remain separate. [Jump conformance](jimmy-jump-motion.md)
adds the exact wide hold comparison and gravity intermediate.
108 saved native observations compare against managed components:25 ground,
35 air,16 facing and32 jump.27 additional native special-contact dispatch cases
establish callback eligibility only.12 air destination writes establish missing
pitch/roll target behavior, which remains a deliberate implementation gate; see
[jimmy-air-motion.md](jimmy-air-motion.md). These are copied-original component
observations, not observed original whole-game play.
### Collision compatibility boundary
The original supported ground apply subtracts250 from live Vy only when actual
support polygon338 is non-null. This is live with the source body and its real
SupportHit. A capsule Grounded flag or floor-height query does not stand in for338.
An attempted mapping from capsule support introduced a repeatable lab stair block
(artifacts/jimmy-movement-audit/lab-route.log, leg1 root4307.62/15.13/351.25).
Removing that mapping restores the entire input-only route
(lab-route-compatible.log), and the two-way underground route also passes. The
capsule retains baseline landing Vy0. This is an explicit compatibility gate,
not an original-body result or a replacement friction/step tuning fix.
### Next coherent implementation chunks
1. **Implemented bounded bridge:** source destination-Euler pitch/roll, retained
apply-owned7EC/7F0, Init/dirty, full authored XYZ and logical-root render basis.
[Air bridge](jimmy-air-bridge.md) records integration and subsequent validation.
The modern local sole correction and final OMT decomposition precision remain
explicit limitations; this does not close full original visual parity.
2. Finish Vec4 previous/live-root publication after callbacks; separate body
integration210 from world-response1D0. Verify successful fence/ladder entry,
animation-end/fallback completion and4255C0 hazard/impulse ownership before
broader Jimmy body admission. The incomplete success-probe experiment is
preserved under ignored artifacts/jimmy-special-contact/success; it is not
conformance evidence or part of the committed dispatcher tool.
3. Recover dynamic pair eligibility/shared context and original support metadata;
replace the lab/outside capsule in an independently verified route chunk.
4. Complete animation/camera/retained helper/cadence comparison on a working
original installation and native Windows. The user's preserved movement
baseline remains authoritative among unverified port alternatives.
### Cross-system regression interpretation
Original AddEuler stores small negative pitch near360; Bubble's test/smoke checks
now compare signed angular direction rather than requiring a negative stored
number. Its fire/throttle and source target sign stay unchanged. Source-supported
Grappler attachment also exposes phase composition: fresh input writes Vy100,
then the ordinary ground controller subtracts250, producing−150; retained late
input writes100 after that publication. The existing ground native positive-Vy
case proves this subtraction. Managed attachment checks now distinguish those
orders. Control loss still runs the body, but a supported root need not move;
its regression checks the physics count/support/ownership instead of inventing
an airborne displacement. Grappler completion/other callbacks are not inferred
from these tests.
The eight-stage Bubble GPU replay exposed the first-step boundary too: a negative
subdegree facing increment can publish canonical360, whose signed remainder is0.
The immediate throttle assertion admits that source result; the sustained managed
case still requires negative lean. Both headless and native1920×1080 SDL Bubble
replays pass all eight phases with zero validation errors. This correction changes
only the smoke predicate, not movement or gadget behavior.
## Verification and package checkpoint
The checks in scripts/test.sh pass across the main run and recorded resumed runs:
320 Python tests,2 HUD extraction checks,43 managed projects, main Vulkan replays,
asset decoder checks, dummy SDL audio/lifecycle and539 original hashes. The main
run first stopped at the obsolete Grappler attachment expectation. Its corrected
13-group suite passes; the remaining script tail then found the first-step Bubble
canonical360 smoke assumption. Corrected headless/native Bubble replays pass, and
the final GPU/save/load/animation/audio tail exits0. The initial failure logs are
preserved. This is a resumed verification record, not a claim that the initial
single script invocation exited successfully.
Artifacts under artifacts/jimmy-movement-audit: full-tests-final.log,
grappler-integrated.log, remaining-tests.log, bubble-final.log, gpu-tail.log and
validation-summary.json. These include36 main Vulkan render completion lines,
all zero validation errors, plus the corrected Bubble replay. Managed component
fixtures compare108 original observations. Both lab/underground navigation and
all source-body gates pass.
Native Linux GTX1070/private Xvfb1920×1080 runs pass walking and powered steering
with shift/right-shift/insert/keypad0/right-mouse, received SDL input and resize.
Each set of five final readbacks is byte-identical. Native Carl/control-phase,
9-stage Grappler and8-stage Bubble pass, with zero Vulkan validation errors.
Logs and captures are in window/. This is port runtime verification; actual
whole-original/native Windows gameplay remains unverified for this checkpoint.
Both self-contained packages identify clean runtime commit
`acd1b287cfdc19e6b5d16c5c10a1d9e9b90c6d41`, source_dirty=no.
Final packaging log: artifacts/jimmy-movement-audit/package-final.log. Each has519
unchanged original asset copies and4271 scene/resource files matching preparation;
package-identity.json records executable/DLL digests and build information.
Standalone Linux from /tmp with DOTNET_ROOT/DOTNET_CLI_HOME/NUGET_PACKAGES unset
passes300-step right-mouse powered flight and240-step Shift walking, received SDL
input and resize, zero validation errors. Both final readbacks match development
byte-for-byte; package-{walk,jetpack}.log and package-readbacks.json retain results.
Post-package audit verifies all539 originals unchanged. Windows cross-publishing
and resource checks do not establish native Windows gameplay. Documentation after
this runtime commit records validation; BUILD-INFO.txt preserves package provenance.