Files
Neutron/docs/gameplay/jimmy-skip-composition.md

8.2 KiB

Original input/controller composition after scene return

2026-11-05. Independent bounded original-code execution for the accepted next-work plan. This addresses whether the port's fresh-versus-retained distinction is an invented implementation artifact or also exists in the known original components. It does not execute original startup, complete Carl activation or all skip logic.

Source identity and recovery

Original Neutron.exe SHA256 f5281076e1a2492f2d61f87aae3f4544b6d0f535ebe33f9fbb6f78710fe1aea8. REA complete instruction Evidence ev_0d64cc58fa8f5573fc8e61e0f2c20e36288099212bd2257d3a3121361fd6c904 provides the bounded source instruction context. The generator verifies that Evidence's subject identity, verifies the authoritative EXE hash, PE-maps every copied range and records source range hashes. Originals are never modified or executed in place. The generated probe runs under the existing isolated Wine32 asset-probe boundary, with no network or original-directory mount.

Executed source and synthetic boundary

Original VA / raw file offset Executed role
431518..431590 /31518..31590 Initial MCA current/destination flatten suffix; actual328→4723F0 and32C→472450 leaves.
430F1D..430F5C /30F1D..30F5C Accepted skip's JIM1 lookup, complete1E00 latch, shared4F83E4 set and audio-stop prefix. Lookup/audio transport are recorded shims; target cleanup and completion are outside this prefix.
475A40..475A4C /75A40..75A4C Actual raw control-pointer grant and removal. Shared skip is explicitly cleared to represent the already-verified DEFAULT field effect; complete DEFAULT is not executed.
438A60..438BA2 /38A60..38BA2 Actual retained-mode dispatch. Its absolute jump-table operand and four table entries are relocated into the private source image.
439900..439E33 /39900..39E33 Actual keyboard/zero-mouse ordinary air input through the real36C→472660 destination writer. The camera tail is replaced by the proper source-function epilogue.
437940..437C3A /37940..37C3A Actual player wrapper including control identity, jump/support dispatch, mode/turn reset and publication. Its40E050 base call is replaced only by actual43A420 facing; unrelated base resource/animation behavior is not simulated.
43A420..43A5CD /3A420..3A5CD Actual facing, actual334→458F90→472480 Euler-add chain; global ready gate forced true.
429910..429963 /29910..29963 Actual landing effect/menu helper with synthetic menu488/noop and invalid sound handle. It does not change mode/turn/Euler.
43915D..4395CF /3951D..395CF Actual neutral ground keyboard suffix after the wrapper has selected mode0; ground camera/device prefix omitted.

The actor is synthetic, supported=true, jump-held=false, ordinary raw gadget-1, zero mouse/analog/keyboard and initial speed0. Complete/game vtables are explicit. Ground/air motion apply and animation actions are recorded no-op resource shims; the known mode/turn writes and facing/input arithmetic execute from original code. The current-action helper represents STOP versus RUN inequality. There is no body, world collision, registry, full derived Jimmy input callback426030, animation engine, renderer, complete MCA update/completion, original startup or full-game execution.

The initial state deliberately models the problematic boundary: retained movement mode1 and stored turn±11.666668, with both current/destination flattened to (0,155,0). After180 uncontrolled player wrappers, those vectors remain upright and mode/turn survive. The composition then grants control and orders input/player callbacks either fresh-input-first or retained-player-first, followed by300 idle callbacks. Each order is run with and without the original skip-prefix writes, and with both turn signs: eight finite cases.

Observed results

All8 cases complete with the required marker and capture wineExit0. Finite post-capture assertions validate the boundary values below.

Order Skip prefix First resumed effect After300 idle callbacks
Input before player Off or on Air input decays turn±11.666668 to±10.000001 and writes target roll∓10.000001. Actual player wrapper then clears mode/turn, after facing has consumed the bad target. Four cases retain approximately±10° bank.
Player before input Off or on Actual supported/released player wrapper clears mode1/turn before input. Mode0 neutral ground suffix preserves upright target. Four cases stay exactly upright.

Representative positive-turn fresh result is bit-for-print equivalent to the existing managed diagnostic: first target roll-10.000001; first player current roll360, mode0/turn0; idle current roll350.000183. Negative-turn result converges to9.99999523. This is original code reproducing the same component composition, not proof that this fresh composition is the normal original playable lifecycle.

With skip enabled, the prefix sets both latch and shared skip to1 and records one stopped sound; mode and turn remain unchanged. Those writes alone therefore do not erase the first stale target in this composition. Their full menu/input consumer and caller timing remain separate questions.

Consequences and unresolved original behavior

The known controller/input contracts agree with the port's first-writer trace; there is no new source evidence for adding a rotation or turn reset at raw grant. The critical question is which callback order and entry state actual original startup and the reported Carl route establish. A fresh allocation partial-order proof must not be substituted for normal New Game execution: original startup/ menu/loading may already have created a retained Jimmy before the first playable lab. Proving that chain would explain upright return without an invented reset. This investigation does not independently establish that startup chain.

Follow-up2026-11-06: Carl/startup audit positively traces MAIN construction and supplies no lab-preload/Jimmy construction in that bounded chain. The new observer also sees only MAIN before the unusable black-screen boundary. Neither result supports switching all fresh scenes to retained order. The broader lifecycle trace keeps actual original Carl entry state/order/delta unresolved instead of adopting the preload hypothesis.

One additional inspected base branch is not a reset justification:40E050 pins current transform to authored transform when game58C CanMove==0, but actual lab JIM1 has serialized CanMove1. Other base resources/callbacks remain omitted from the composition and must be tracked if startup/order does not explain the user's whole-original observation.

Artifacts and reproduction

  • tools/binary/jimmy_skip_composition.py and _probe.c: read-only source guards, relocation generator and bounded synthetic composition.
  • artifacts/jimmy-skip-composition/build/source.json: ten original range hashes, PE offsets and relocation operands.
  • artifacts/jimmy-skip-composition/native-observations.json: all eight cases, per-boundary fields and source/harness/executable/log/capture provenance hashes.
  • probe/wine.log and probe/capture.json: successful marker and wineExit0.

From GamePort:

python3 tools/binary/jimmy_skip_composition.py ../Original \
  --output artifacts/jimmy-skip-composition/build \
  --rea artifacts/jimmy-movement-audit/rea-original-disassembly.json
i686-w64-mingw32-gcc -O0 -Wall -Wextra -Wl,--image-base,0x10000000 \
  -I tools/binary artifacts/jimmy-skip-composition/build/jimmy_skip_composition_probe.c \
  -o artifacts/jimmy-skip-composition/build/probe.exe
python3 tools/recon/reference_capture.py --asset-probe \
  --probe-executable artifacts/jimmy-skip-composition/build/probe.exe \
  --probe-success JIMMY_SKIP_COMPOSITION_COMPLETE --seconds 12 \
  --output artifacts/jimmy-skip-composition/probe

Harness corrections preceded the successful run: initial Wine prefix creation outlasted its first capture, the absolute dispatch jump table needed relocation, and the skip-prefix exit initially preceded its cdecl argument cleanup. These failures were probe construction gaps, not original gameplay observations. The final compiled probe has no warnings; only the final success marker/exit and validated eight-case outputs establish the observations. No production/runtime/ regression-test/build-script change or commit is made by this investigator.