Fix lab cutscene control, landing steering, clipping and objective timeout

This commit is contained in:
UnixNight committed 2026-10-25 01:18:08 -04:00
1 parent 976e677019
commit ca08c468ba
33 files changed
+871 -64

No files matched your search

+18 -2
View File
@@ -46,8 +46,9 @@ velocityXZ = forward(yaw)*speed + right(yaw)*sideSpeed
Source friction is0.8. Air sideSpeed lacks the additional ground dt factor;
ground437CDx uses (turnRate*(1-friction)*speed/45+keyboardSide)*dt instead.
The port now preserves the shared turn-rate field through landing and clears it
on scene reset. Air input's destination yaw is current yaw (439E2B), so the ground
Correction:437B9E..437BAC clears the shared turn field on grounded/released
input. Its receiver is the game interface, so interface+614 is complete+6D4.
The port now clears it at this boundary and on scene reset. Air input's destination yaw is current yaw (439E2B), so the ground
keyboard target offset does not override its direct yaw update.
AirSteering.cs translates this keyboard subset. LabSimulation calls it while
@@ -83,3 +84,18 @@ with identical403.88 powered travel and grounded position. Linux reports zero
validation errors; Wine validation is disabled. Logs: artifacts/package-jetpack-steering.log
and artifacts/windows-jetpack-steering-package/wine.log. Native Windows remains
unverified by the agent. All539 original hashes still match.
## Ground-turn cancellation correction
User feedback after the first air-input fix identified intermittent slow turning
with all jump buttons. The previous integration incorrectly retained TurnRate
on the ground: an air turn of-30 could cancel most of the opposing+37deg/sec
walking turn. The original outer reset437B9E..437BAC clears jump state, mode and
interface+614 (complete+6D4). The port now performs that grounded/released-jump
reset before ground steering. A17th movement group reproduces opposing air turns
landing into left/right ground input and requires the full source turn rate.
A third source-byte regression guards all20 reset bytes and pointer displacement.
All five SDL bindings (left/right Shift, Insert, keypad0, right mouse) also pass
identical300-frame powered steering replays, each403.88 travel. No button-specific
turn multiplier is introduced. Original global update phase order remains open.
+103
View File
@@ -0,0 +1,103 @@
# Original geometry clipping overrides
User testing reported a visible distance clipping plane during gameplay which
disappears in orbit inspection. The port selected10000 only forLV1B and3200 for
every other level.3200 is the OMediaLayer constructor default, not the original
gameplay setting for those levels. The following static recovery reads original
Neutron.exe only; original-game visual comparison remains separate.
Neutron.exe SHA-256: f5281076e1a2492f2d61f87aae3f4544b6d0f535ebe33f9fbb6f78710fe1aea8.
EXE text file offsets equal VA minus0x400000 in these ranges.
## Original setup and decisive regression
OMediaLayer ctor DLL1004D340 initializes near20, far3200, FOV0xAAA OMT angle
units (59.9853515625 degrees). Game geometry layers are globals509A30/509A3C;
their far field is+0xC0. Original level complete-object RTTI/vtables identify
initializer virtualB4, whose stores below override those constructor values.
The near20 and FOV are retained. Sky layer509A38 has its own1000000 far plane;
this helper does not introduce missing sky rendering or sky-specific states.
Outside LEV1 initializer450580 writes float28000 at45086F to geometry layer
509A3C. VR04 initializer456AD0 writes50000 at456C04/456C4A to both geometry
layers. A gameplay3200 far plane therefore clips valid original terrain and
props in both scenes. Orbit used scene-size-derived far=radius*20 and masked
the wrong gameplay range rather than correcting the scene setting.
Original level1 geometry spansX[-11080.31,34316.65],Y[-546.06,19685.04],
Z[-18442.92,26509.07]; VR04 spansX[-6273.61,4535.86],Y[-2.03,7815.47],
Z[-7364.04,2746.19]. Geometry bounds corroborate the impact but are not used
as a substitute for the recovered original clipping values.
## Recovered initial geometry-layer far limits
All35 registered level initializers were inspected. This includes3B/E records
without establishing their normal-gameplay use. Values are world units. An
unwritten layer retains its constructor3200.
| Original class / GAM code | Initializer VA | Layer509A30 far | Layer509A3C far | Store VAs |
| --- | --- | --- | --- | --- |
|CLevel01AGame / LV1A|0x44f0e0|5000|5000|0x44f210, 0x44f256|
|CLevel01BGame / LV1B|0x44f4b0|10000|10000|0x44f5a5, 0x44f5b1|
|CLevel01CGame / LV1C|0x44f7e0|10000|10000|0x44f896, 0x44f8a2|
|CLevel01DGame / LV1D|0x44fb00|120000|120000|0x44fc35, 0x44fc41|
|CLevel01EGame / LV1E|0x44fe30|120000|120000|0x44ff65, 0x44ff71|
|CLevel01FGame / LV1F|0x450160|15000|15000|0x4502da, 0x4502f2|
|CLevel01Game / LEV1|0x450580|3200|28000|0x45086f|
|CLevel02AGame / LV2A|0x450ab0|3200|3500|0x450c21|
|CLevel02BGame / LV2B|0x450e30|50000|50000|0x45112d, 0x451173|
|CLevel02Game / LEV2|0x4513e0|3200|50000|0x4516b2|
|CLevel03AGame / LV3A|0x451890|3200|80000|0x451a3e|
|CLevel03BGame / LV3B|0x451c50|40000|40000|0x451e06, 0x451e29|
|CLevel03CGame / LV3C|0x452030|40000|40000|0x4521e6, 0x452209|
|CLevel03DGame / LV3D|0x452410|40000|40000|0x4525c6, 0x4525e9|
|CLevel03EGame / LV3E|0x4527f0|40000|40000|0x4529a6, 0x4529c9|
|CLevel03Game / LEV3|0x452c20|40000|40000|0x452dd7, 0x452dfa|
|CLevel04AGame / LV4A|0x453060|10000|10000|0x45320a, 0x45322d|
|CLevel04BGame / LV4B|0x4534c0|7200|7200|0x4535bd, 0x45364c|
|CLevel04CGame / LV4C|0x453b10|30000|30000|0x453dfb, 0x453e1e|
|CLevel04DGame / LV4D|0x454070|10000|10000|0x45421a, 0x45423d|
|CLevel04Game / LEV4|0x4544a0|30000|30000|0x4546d6, 0x4546f9|
|CLevel05AGame / LV5A|0x454960|16000|16000|0x454a94, 0x454ad6|
|CLevel05BGame / LV5B|0x454ce0|16000|16000|0x454e14, 0x454e56|
|CLevel05Game / LEV5|0x4550b0|12000|12000|0x455242, 0x455288|
|CLevel06AGame / LV6A|0x4554e0|4000|4000|0x455614, 0x455656|
|CLevel06Game / LEV6|0x4558b0|4000|4000|0x4559e4, 0x455a26|
|CLevel07Game / LEV7|0x455c80|10000|10000|0x455dff, 0x455e41|
|CLevelVR01 / VR01|0x456050|50000|50000|0x456184, 0x4561ca|
|CLevelVR02 / VR02|0x4563d0|50000|50000|0x456504, 0x45654a|
|CLevelVR03 / VR03|0x456750|50000|50000|0x456884, 0x4568ca|
|CLevelVR04 / VR04|0x456ad0|50000|50000|0x456c04, 0x456c4a|
|CLevelVR05 / VR05|0x456e50|50000|50000|0x456f84, 0x456fca|
|CLevelVR06 / VR06|0x4571d0|50000|50000|0x457304, 0x45734a|
|CLevelVR07 / VR07|0x457550|50000|50000|0x457684, 0x4576ca|
|CLevelVR08 / VR08|0x4578d0|50000|50000|0x457a04, 0x457a4a|
## Implementation boundary
Application/SceneProjection.cs supplies FarPlane(code) and CreateGameplay(code,
aspect). CreateGameplay returns the original positive-Z,0..1 depth perspective;
the existing caller applies Vulkan viewport Y flip. Code lookup is case-insensitive.
Unknown/model-only contexts retain the original generic3200 default.
The current port merges both geometry layers into one draw collection. Where the
original layers differ, it uses their larger far limit. That is an explicit port
policy which prevents truncating the layer with the longer original range; exact
per-layer clipping/masks still require recovered render-layer assignments.
LV4B is a known dynamic exception: initial7200 is stored4535BD/45364C. Task
callback453690 sets flag+530 when task value1; update4536C0 then animates far
using12000+19000*timer at453780..453792/453811..453829, finally setting22000
at45388A..45389E. The standalone helper returns the initial7200 only; task-
driven fog/color/clip animation is not implemented by this static mapping.
## Verification
ProjectionTests guard exact original outside and VR04 layer-store instructions,
map every supplied GAM code, check depth0/1 endpoints for all ranges, preserve
generic fallback and reject invalid aspect ratios. A real outside source camera
and source geometry regression counts vertices which fit the original28000
frustum but were cut by the old3200 far plane:3844 vertices are recovered in
the real LEV1/level1 source view. All three managed groups pass.
Runtime GPU/readback integration
belongs to primary; these checks do not by themselves establish rendered fidelity.
+32
View File
@@ -0,0 +1,32 @@
# Carl's lab conversation control handoff
User testing found Jimmy remained controllable during the lab's ending Carl
conversation. Static original evidence: LV1B neutron1a is a five-shot3MCA with
PlayerControlled=JIM1 and NextTrigger=carlwalk. The earlier interpreter applied
PlayerControlled at activation, prematurely granting the saved completion target.
Original MCA activation431280 chooses the first valid CameraTarget. When an
actor currently owns control,431502..43159E performs actor preparation, then
43159E..4315A7 calls game virtual118(NULL). The original game setter475A40 simply
assigns global controlled entity5099E4. The multi-camera completion4311F0..431228
reads PlayerControlled and applies its none/NULL/named-target semantics at the end.
JIM1 in neutron1a therefore restores Jimmy after the conversation, not before it.
SceneDirector.StartMulti no longer applies the completion field. Each valid MCA
shot clears ControlledTag, and FinishMulti retains the recovered restoration.
Program's existing simulation gate therefore suppresses walking/turning/jumping
throughout the conversation. Camera, actor/patrol and dialogue clocks continue;
Escape/Space skips while preserving authored follow-up actions.
DirectorTests now execute the real neutron1a with natural and skipped completion,
assert control is absent through every active shot, restored afterward, and guard
the exact original43159E instruction sequence. Interaction tests retain the
pre-dialogue score150/SCENE60 side effects. A60-frame headless runtime check
(--smoke-carl-cutscene) holds move/turn/jump requests and records zero simulation
steps with active neutron1a and zero Vulkan validation errors. This source/contact
smoke does not replace live navigation or original-game comparison.
The user separately confirms lab -> tutorial -> lab -> outside is reachable in
their tested build. Platform and exact checkpoint-return-position fidelity were
not specified by that report; do not infer complete native-platform/original
fidelity from transition reachability alone.
+107
View File
@@ -0,0 +1,107 @@
# Objective pad timeout, slide and O key
Read-only Neutron.exe recovery plus managed regression checks. Original text VAs
below correspond to file offsets VA-0x400000. This fixes the earlier stationary
HUD approximation; original-game/runtime visual comparison remains separate.
## Timer and phases
Objective selector4F8194 is distinct from animation timer4F8188.406F50 ignores
nonpositive selectors; for a positive selector it clears44 flags, stores selector,
resets timer0 at406F76 and enables the selected flag. Initial UI4018A8 and Jimmy
PostLoad423EC6 set timer20, so merely loading a saved positive selector does not
prove that its popup should initially appear.
4043D9 checks current controlled entity5099E4 againstNULL. Only a non-null entity
advances the timer by elapsed seconds at4043E5..4043F1 and invokes407070 through
virtual4D0. No controlled entity takes40447A..4044BD, hiding objective art and
leaving the timer unchanged. The test is for any controlled entity, not necessarily
Jimmy; cutscene control policy must be supplied separately by the caller.
407070 uses these phases (source floats retain their original widths):
| Timer t | Pad | Objective art/digit | Pad vertical offset from stationary layout |
| --- | --- | --- | --- |
| t<0.7 | Visible | Hidden | (1-t*1.4285714626312256f)*100 |
| 0.7<=t<7 | Visible | Visible |0|
|7<=t<8.7 | Visible | Hidden |(t-7)*1.4285714626312256f*100|
|t>=8.7 | Hidden | Hidden |No draw|
Entry threshold0.7 is48D974; reciprocal scale is48D970; distance100 is48D96C.
Stationary end is signed integer7 at4EC498. Exit draw window ends at7+1.7f,
with1.7 at48D940. The longer exit draw window means the pad can already leave
the viewport before the branch finally stops drawing it.
Pad top-left is x384,y=380+self[4EC]+slideOffset. Stationary art is
x384,y=448+self[4EC] (plus15 per visible row), digit x393,y=450+self[4EC].
Entry4070FD..407138 draws only the pad. Stationary4071BA..407247 shows selected
flags, digit and pad. Exit407266..40727E hides the44 objective art elements,
then407280..4072C4 draws the departing pad. At equality8.7 the40725F test exits.
Entry sets self[4E8] and emits effect216 once at4070BA..4070CA. Stationary
4071BA clears that latch. Resetting the timer while still entering therefore
does not necessarily replay the sound; ObjectivePanel retains this distinction.
## O opens/closes the pad, separate from gadget layout
407071 reads509887, which is virtual key'O' (4F) plus the recovered keyboard
array509838. Controller flag509AC9 joins that condition.4F7C30 is its edge latch:
no input clears the latch at40708F; a newly held input sets it at40715B.
On that edge40714F..407198:
- If timer is nonnegative and less than7, set it to7 and begin slide-out.
- Otherwise reset it to0 and begin slide-in.
- Holding the input does not repeat the toggle. Release then press creates a new edge.
The previous port used O to toggle layout offset-69/0. Original helpers406590
and406650 do change that offset and recreate/hide other gadget UI resources,
but407070's O branch does not call those helpers. O must control the timer instead.
The current objective consumer retains layoutOffset=-69; other gadget layout
transitions remain a separate system.
## Practice/pause policy
While a practice modal is active (excluding source type12),4044D7..404504 sets
the timer to7+1.7 and hides pad and art. No404D97 resets the timer to0, so dismissing
practice can explicitly reopen the current objective. This is distinct from
pausing the timer because no controlled actor exists. Focus/inspection/global
UI-pause scheduling is a caller policy until the complete original UI loop is
recovered; the component accepts explicit elapsed seconds and gating arguments.
## Component and integration API
`Game/ObjectivePanel.cs` owns only this presentation state:
```
var panel = new ObjectivePanel(currentSelector, layoutOffset: -69);
interactions.ObjectiveChanged += panel.Select;
// Each active UI tick; pass the actual held O state only when UI input is allowed.
panel.Key(oHeld);
panel.Step(seconds, hasControlledActor, modalActive);
// Pad and content need independent draw visibility.
panel.PadVisible; panel.ContentVisible; panel.YOffset;
// On practice No, matching404D97:
panel.Show();
```
`Select(positive)` resets timing even when selector is unchanged. Nonpositive
assignments are the recovered no-op. The current renderer has one selected lab
objective instead of all44 independent flags; selector0 deliberately shows no
invented art/digit. `SoundEffect` emits216 at the source entry latch. Constructor
layout bounds and selectors above43 reject malformed caller input.
## Verification
`source scripts/env.sh`, then:
```
dotnet run --project tests/GamePort.ObjectiveTests -- ../Original
```
Six groups pass: exact original constants/instructions; entry/hold/exit positions
and boundary equality; O edge/hold/reopen/layout separation; repeated selector
assignments and0 no-art; controlled-entity/modal/pause lifecycle; source sound
latch plus malformed inputs. The tests compare original Neutron.exe bytes at
seven decisive timer/input/visibility sites. No original file is modified or
executed. Renderer composition/native runtime tests belong to primary integration.
+6 -6
View File
@@ -26,16 +26,16 @@ y=`448 + self[4EC] + visibleRow*15`. `407070` renders the pad at x384,
y=`380+self[4EC]` during its stationary phase and prints selector `%d`
(`4EC794`) at (393,450+offset), font tag SMAL (`534D414C`). `401F9D`
initializes offset=-69; `406590` changes -69 to0, while `406650` changes0 to-69.
These source paths toggle vertical layout and other gadget UI, not a simple
objective visibility bit. The pad extends beyond the640x480 reference viewport;
These source paths toggle vertical layout and other gadget UI. They are separate
from O, whose recovered timer toggle is documented in objective-timing.md. The pad extends beyond the640x480 reference viewport;
retain clipping rather than repositioning it to fit.
`406F50(index,flag)` only accepts index>0: it clears44 dword flags at4F7D00,
sets selector4F8194, resets animation timer4F8188, then enables the selected
entry. It does not map selector0 to objective1. The source has sliding/timed
objective presentation; the consumer currently renders its stationary geometry
under caller-controlled visibility. Animation and exact source visibility timing
are not yet implemented.
entry. It does not map selector0 to objective1. The source has sliding/timed objective presentation, now implemented by
Game/ObjectivePanel.cs:0.7-second entry, stationary until7 seconds, exit until8.7.
Pad and content have independent visibility. O toggles this timer.
See objective-timing.md for exact source rules and regression evidence.
`402D70/402CC0` convert positions using windowWidth/640 and windowHeight/480;
constants at file8D40C=1/640 and8D408=1/480. The port quads use NDC
+13
View File
@@ -122,3 +122,16 @@ walking input. Mode1 logs then falls through to its real handler; it is not a
no-input state. LabSimulation preserves the turn-rate field, uses recovered air
coasting/braking and applies the distinct ground/air lateral formulas. Mouse,
analog, full lean/camera and original global phase order remain incomplete.
MCA PlayerControlled is a completion target. Valid multi-camera shots clear active
control; the player simulation is gated until completion/skip restores it. The
normal grounded/released controller reset clears the air turn field before ground
steering, preventing prior air turns from cancelling keyboard turns after landing.
ObjectivePanel owns source elapsed time, entry/stationary/exit phases and O edge
latching; SceneHud renders its pad and text/digit visibility independently. Practice
hides the popup and No resets timing. UI clock remains explicit across control loss,
focus/inspection and fixed simulation ticks. SceneProjection maps all35 original
level geometry-layer far overrides instead of a universal3200. The merged renderer
uses the larger layer limit; per-layer masks and dynamic LV4B fog/clipping remain open.
+15 -7
View File
@@ -16,7 +16,7 @@ separate evidence categories. **The game is not fully reconstructed/playable yet
| Windows walking package under Wine64 | Preliminary240-step window pass | SDL arrows/Shift, walking/jump/landing and800x600 resize; matches Linux movement statistics; validation disabled |
| Linux self-contained package | Native headless/window verified |12s story without SDK/source/Original mounted; GTX1070 practice pause/N/resume/resize, zero validation errors |
| Vulkan1.3 lab | llvmpipe + GTX1070 | Real indexed/textured/depth frame/readback; validation enabled, zero errors |
| Original camera/projection | Data/math integrated | GAM camera, recovered FOV/clips; startup/follow behavior not compared |
| Original camera/projection | Source mapping/math + outside GPU before/after |All35 level far overrides; outside28000/tutorial50000; merged layer maximum is a port policy, per-layer masks/dynamic fog incomplete |
| Jimmy normal texture / morph | Native preview verified | Correct atlas slot0, base topology/UV retained; state speed/loop unverified |
| Standing preview placement | Corrected / Linux tested | +38.503Y from actual visible floor/base soles; exact original actor transform unresolved |
| Idle/run/jump/scratch preview | Linux keyboard + GPU tests | Default idle samples identical; 1/2/3/4 select clips; run image changes across frames |
@@ -27,13 +27,13 @@ separate evidence categories. **The game is not fully reconstructed/playable yet
| Mouse/relative input/controller | Incomplete | SDL entry point declaration alone is not behavior |
| Audio stream | Linux dummy device | Original11025Hz mono WAV consumed10752bytes/500ms; speakers untested |
| Physics fragments/RNG | Standalone C# tests | Recovered arithmetic; controller/jump/collision state incomplete |
| Walking/jump/air controller |16 movement groups + GPU replay pass |Corrected mode1 fall-through to source air handler: acceleration/braking/turning during jumps and bursts; phase/collision/lean remain provisional |
| Walking/jump/air controller |17 movement groups + GPU replay pass |Corrected mode1 fall-through to source air handler: acceleration/braking/turning during jumps and bursts; phase/collision/lean remain provisional |
| Static source collision | 11 deterministic groups pass | Capsule sweep/slide, walls/ceilings/slopes/ledges; all2869 lab triangles retained; masks/dimensions provisional |
| Simulation cadence |16 movement groups pass | Fixed60Hz independent of60/120/144Hz rendering, ten-minute finite run; original cadence unresolved |
| Simulation cadence |17 movement groups pass | Fixed60Hz independent of60/120/144Hz rendering, ten-minute finite run; original cadence unresolved |
| Walking camera and animation | Linux headless/window pass | Original five motion clips, recovered eye180/target99; pivot, timing and obstruction solver provisional |
| Walk SDL input/resize | Earlier GTX1070/Xvfb240-step pass |Historical pre-air-fix travel1250.12; current corrected managed/headless replay1264.55. Current powered SDL check is below |
| Walk SDL input/resize | GTX1070/Xvfb input replay pass |Current corrected managed/headless replay1252.09; all five jump bindings produce identical300-frame powered replays,403.88 travel; zero validation errors |
| TSK saves/task gates | All12 originals byte-round-trip | Bounded parse/write, task/items/score/checkpoint CLI resume; full game-state/menu save incomplete |
| Lab CAM/MCA/AI interpreter | Source sequence tests + Linux GPU pass |16 speech cues, recovered camera formulas, skip/actions/control restoration; no original comparison |
| Lab CAM/MCA/AI interpreter | Source sequence/Carl lock tests + Linux GPU pass |16 intro cues; neutron1a locks move/turn/jump for all active MCA shots, restores JIM1 at completion/skip; no original comparison |
| Dynamic actors/sprites/models | Linux GPU pass |Carl/Goddard/Jimmy morphs, source Canv timing and9 original lab props; billboard/alpha, pivots and playback timing remain provisional |
| Scene music/speech | Linux dummy-device pass |CAM stops music before speech; full lab chain stays silent, DEFAULT restarts on completion/skip; reset tested. Packaged Linux/Wine64 lifecycle checks pass; speakers/latency/native Windows unverified |
| VR01 star/trophy contacts | Source-data tests pass |5-point stars,300 trophy, acquired flags and contact-enter semantics; distance/exclusions/box approximation |
@@ -43,14 +43,14 @@ separate evidence categories. **The game is not fully reconstructed/playable yet
| Patrol/support |10 patrol +13 support groups pass |Source phases/profiles and8 DLL magnitude refs; Carl sensing, Goddard follow and full wall response incomplete |
| Buttons/doors | Source state tests and GPU props pass |Cooldown, hold/reversal, original model transforms; source broadphase/box order with modern narrowphase |
| Dynamic collision |10 managed groups pass |Includes actual closed/open lab bars; full source solver/activation exclusions incomplete |
| Lab objectives/practice | Art/data/managed/headless + native Linux window pass |Exact source coordinates, modal pause, Y/N/pointer rules; N resumes and resize800x600 onGTX1070, zero errors; slide/highlight incomplete |
| Lab objectives/practice | Source timer/O tests + managed/headless/SDL pass |Original popup slide/timeout and O latch integrated; pad/content separate; modal pause/No reset; practice highlights/controller still incomplete |
| Jetpack tutorial |14 vertical groups +16 movement groups +300-step VR04 GPU pass |Powered horizontal travel403.88, turn/brake/release and landing; rise418.98, zero validation errors; phase/shadow/collision/lean approximations |
| Powered SDL/package steering |Linux GTX1070 and preliminary Wine64 window pass |300 frames, Up/Left/Down/Shift received, resize800x600, travel403.88/landing; Linux zero validation errors, Wine validation disabled; native Windows pending |
| Default effects bank |251 exact RIFF exports, SDL dummy pass |Pickup/button/door sounds and owner-loop stop; positional/volume/channel policy incomplete; MPEG-in-RIFF bird1/id162 unsupported by SDL WAV |
| Complete scene/gameplay loop | Incomplete |Lab navigation, exact collision/pivot/cadence and original reference comparison still required |
| Menus/AI/combat/progression/saves | Incomplete | Required future work |
| Original execution | Attempted under isolated Wine11 | Window created, GL/Vulkan rendering unusable; no reference comparison |
| Native Windows functionality | Loading/rendering user-confirmed | Current story/tutorial/input/audio/resize still unverified natively |
| Native Windows functionality | User confirms lab/tutorial/lab/outside and corrected build works |User built/tested Windows; comprehensive native input/audio/resize/driver coverage and original comparison remain unverified by the agent |
Generated evidence: artifacts/full-integration-test.log, practice-window.log,
practice-transition.log, jetpack-runtime.log, final-tests.log, test-level1b.ppm, test-player.ppm,
@@ -59,6 +59,14 @@ The lab/player images have been visually inspected; they do not establish origin
visual equivalence. Windows cross-publishing must not be represented as Windows
runtime validation. Package scope and prerequisites are in infrastructure.md.
Current four-fix regression evidence: artifacts/four-bugs-full-tests.log (complete
suite and539 preserved hashes), jump-binding-parity.json, outside-fixed-clip.png,
package-objective-window.log and package-carl-cutscene.log. Both platforms publish
successfully; packaged Linux popup/Carl checks pass with zero Vulkan validation
errors. The already-started Wine64 popup/O/resize check passes with validation
disabled. Further package checks were skipped at the user's request after they
reported their own Windows build works. No new agent native Windows claim is made.
Known visual/behavior limits include incomplete lighting/fog/blend/culling/material
mapping, inferred model/root pivot, provisional collision/cadence/animation timing,
close-wall camera clipping near Jimmy, incomplete class-specific triggers/interactions,
+6
View File
@@ -63,3 +63,9 @@
15. **Default recovered scene play.** Start the integrated lab story/interactions by
default; retain --inspect and deterministic smoke modes. Passing direct-contact
tests does not establish navigation, original fidelity or full-game completion.
16. **Use recovered control and UI lifetimes.** MCA PlayerControlled restores its
target at completion; active valid shots clear control. O changes the objective
animation timer rather than gadget layout. Reset air-turn state at the recovered
grounded/released branch. Use per-level geometry-layer far limits; the current
merged renderer uses their maximum until separate layer masks are implemented.
Source evidence and regression limits are recorded in the focused analysis docs.
+38 -4
View File
@@ -1,9 +1,43 @@
# Progress — 2026-10-24
# Progress — 2026-10-25
**The full reconstruction is incomplete.** The current implementation extends the
user-verified Windows rendering checkpoint into an interactive original lab and
jet pack tutorial flow. This is not a verified faithful complete scene or release.
## User-reported lab regressions
The user confirms lab -> tutorial -> lab -> outside is reachable. Four follow-up
bugs are now implemented with source evidence and targeted regressions:
- Carl's neutron1a MCA no longer grants its JIM1 completion target at activation.
Active shots clear control; natural/skip completion restores it. Runtime holds
move/turn/jump for 60 frames and records zero player simulation steps.
- Grounded/released jump resets the source turn field (interface+614=complete+6D4).
Residual air turn previously opposed subsequent ground steering. 17 movement
groups pass, including landing into both opposite directions. All five jump
bindings pass identical SDL powered replays; no button-specific turn scale.
- Gameplay clipping now uses recovered original level overrides: outside28000,
tutorials50000, all35 level mappings. The old universal3200 clipped valid distant
scenery; actual outside GPU before/after readbacks confirm its recovery.
- Objectives have source 0.7-second entry, hold until 7, exit until 8.7. O opens/closes
via an edge-latched timer instead of changing layout. Modal/control gating and
source sound216 are integrated; separate pad/content draw checks pass.
The complete scripts/test.sh run passes, including 39 Python checks, managed
regressions, scene/GPU/audio checks and all 539 unchanged original hashes
(artifacts/four-bugs-full-tests.log). Both self-contained packages are rebuilt.
Packaged Linux popup/O/resize and 60-frame Carl control checks pass on GTX1070
with zero Vulkan validation errors. The already-started preliminary Wine64 popup
check also completed successfully (validation disabled). Further package checks
were skipped at the user's request: the user reports that their own Windows build
works with these corrections. That report establishes user verification, not an
agent-run native Windows comparison or complete input/audio/driver coverage.
Reports: analysis/carl-cutscene-control.md, air-steering.md, camera-clipping.md and
objective-timing.md. Original comparison, per-layer rendering/dynamic fog and full gameplay
remain incomplete. User loop reachability is distinct from source-faithful
checkpoint-position restoration and agent native Windows validation.
## Jetpack steering correction
User testing exposed frozen airborne steering. The earlier recovery incorrectly
@@ -55,7 +89,7 @@ navigation coverage without proving original collision/cadence fidelity.
- The pickup's internal tag `shrinkray` names the lab jet pack, proven by original
GET JET PACK art and powered-jump inventory checks. It is not a shrink-ray weapon.
- Original objective pad, three lab objective images, digits and PracticeYesNo
are rendered at source640x480 coordinates. O changes stationary panel offset.
are rendered at source640x480 coordinates. O opens/closes the timed popup.
Practice pauses the simulation; Y enters VR04/PHONEBOOTH and N resumes.
Return context is current GAM/latest STRT (LV1B/STARTEXP for NewGame), separate
from the still-unrecovered runtime checkpoint-root restoration.
@@ -89,8 +123,8 @@ confidence and approximation boundaries are retained there.
Managed checks cover source data, director/actions/cameras, task/interactions,
patrol, recovered support, moving-box collision, jetpack states and HUD composition.
Python currently has38 top-level regression tests plus2 targeted UI extraction tests.
Existing11 collision and16 movement groups pass;14 jetpack groups pass. Eight
Python currently has39 top-level regression tests plus2 targeted UI extraction tests.
Existing11 collision and17 movement groups pass;14 jetpack groups pass. Eight
quick-magnitude float results are checked against an isolated original DLL probe.
Original mode2 standing/jump animation probes confirm zero dynamic-point offsets
for the tested clips. This is DLL evidence, not observed original game behavior.
+3
View File
@@ -16,6 +16,9 @@ probes and modern runtime checks are recorded separately.
- `analysis/dynamic-collision.md`: source3DUD broadphase/triangle order and modern response boundary.
- `analysis/buttons-doors.md`: button contact/cooldown and moving-bar state machine.
- `analysis/ui.md`: original objective/practice canvases, coordinates, input and return strings.
- `analysis/carl-cutscene-control.md`: MCA activation locks control; serialized target restores it at completion.
- `analysis/objective-timing.md`: source popup timer/slide/O latch and modal clock behavior.
- `analysis/camera-clipping.md`: per-level original near/far geometry clipping and merged-layer limits.
- `analysis/air-steering.md`: corrected mode1 fall-through, airborne keyboard handler and source regression evidence.
- `analysis/jetpack.md`: corrected lab pickup identity, powered-jump/effect/energy phases.
- `analysis/sound-effects.md`: default bank identity, preserved wave payloads and loop lifecycle.
+11 -7
View File
@@ -23,16 +23,20 @@ compile, runtime and original-fidelity verification are separate.
- [x] Powered-jump state/effect/energy arithmetic, original FLY/fire morphs and GPU replay.
- [x] Correct mode1 input dispatch and source airborne steering for jumps/bursts;
Up acceleration, Down braking, Left/Right turn and powered travel regressions.
- [x] Carl conversation control lock, grounded air-turn reset, original level far
planes and timed objective/O behavior. Full suite passes; both packages rebuilt;
user verifies their own Windows build works. Detailed evidence: progress.md.
## Immediate continuation
1. Integration milestone c929f94 and the subsequent cutscene music/navigation
correction are implemented/tested and republished. Preserve native Windows
story/input/audio as pending; source user feedback is distinct from agent Wine
checks. Continue original physics/audio/UI recovery and lab underground traversal.
1. Lab regressions are corrected and republished, with user verification of their
own Windows build. Keep comprehensive native Windows driver/audio/input
coverage separate from that report and preliminary Wine checks. Continue
original physics/audio/UI recovery and lab underground traversal.
2. Managed input-only lab navigation now reaches the outside LOAD with live source
contacts and NPC updates. Keep the direct-contact smoke distinguished from this
replay. Extend coverage to underground button/door traversal and runtime
replay. User confirms lab/tutorial/lab/outside is reachable. Extend coverage to
underground button/door traversal and runtime
reference comparison. Recover runtime42AF00 checkpoint restoration on
VR return separately from saved STRT/RETURN strings and unknown TSK StoredVector.
3. Recover47BD80 collision response/479D20 movement phase and source object exclusions.
@@ -43,8 +47,8 @@ compile, runtime and original-fidelity verification are separate.
policies, jetpack sound callbacks and MPEG-in-RIFF bird1/id162 decoding. Default
SoundEffects.omt identity and pickup/button/door stream lifecycle are integrated;
physical/native Windows output and original timing remain unverified.
5. Complete objective slide/timing, practice highlights/cursor/controller handling,
energy/gadget UI. Stationary geometry and keyboard/pointer choices are recovered.
5. Complete practice highlights/cursor/controller handling and energy/gadget UI.
Objective slide/timeout/O behavior is now recovered and integrated.
6. Recover Carl sensing409CC0, Goddard follow-state4, actor class defaults and full
morph state/playback-end synchronization. Generic patrol states alone are not AI.
7. Recover original global cadence/object phase order, mouse steering/relative input,