# Compatibility and validation 2026-11-07 normal-callback checkpoint:30 complete emitted original/C# fixture snapshots match exact float32 values, raw angle arguments/results and getter order. The inherited ceiling/reflection/angular component remains outside live profile wiring. Jimmy no-effect paths and Soldier174/Jimmy204 record-only endpoints are explicitly bounded; no full derived policy is implied.352 Python checks and targeted sphere/static/body/ownership/pair/context/reclip checks pass. Linux Debug build has zero warnings/errors; Windows x64 self-contained cross- publish succeeds.539 originals and both924ad2f packaged DLLs/BUILD-INFO are unchanged. No new graphics/native SDL, whole-original or Windows execution is claimed; integrated runtime evidence remains bf876e9. Invalid partial captures from the corrected relocation scan are excluded. See [normal callbacks](checkpoint/actor-normal-contacts.md) and local artifacts/actor-contact/validation.json for source/capture/build identities. 2026-11-07 populated-reclip correction:14 complete original/C# observations and12 kernel steps match exact float32 values after source-store arithmetic corrections. Initial node-visit, edge-projection and ray-hit failures are retained. No epsilon or near-zero reset is added. `scripts/test.sh` exits0:348 Python checks,2 HUD Python checks, all managed projects,36 Vulkan/render checks with zero validation errors, dummy-device audio lifecycle and539 unchanged original files. Debug build has zero warnings/errors. Native Linux1920×1080 SDL Carl-control and VR04 jetpack/ resize replays also pass with zero validation errors; first neutral input and300 idle ticks stay upright. Windows x64 self-contained cross-publish succeeds to a separate artifact directory; no new Windows execution is claimed. Existing Linux/ Windows packaged DLLs and BUILD-INFO remain the verified924ad2f baseline. Dynamic composition still uses synthetic actors/callbacks/grid and is not live pair admission. Double intermediates reproduce these cases, not all x87 inputs. See [populated reclip](checkpoint/dynamic-reclip.md), with final logs/identities under artifacts/dynamic-reclip/validation.json and its referenced artifacts. 2026-11-07 dynamic-context checkpoint:17 copied-original mode2/static reclip observations match the C# test composition in full, including XYZW, shared nearest/hit/old polygon, callback/setter/pass counts and mode. All12 instrumented kernel triangle/normal arguments also match. A retained diagnostic intentionally fails with construction winding despite the same final position. The source static grid is empty, transforms/callbacks are synthetic and mode starts0; nonempty/reentrant/live actor behavior is not established. All345 Python tests, 17 fixed context cases,11 static groups,10 dynamic groups and28 pair cases pass. 539 originals and both924ad2f packaged DLLs/BUILD-INFO are unchanged. No runtime code, package refresh, new GPU run or native Windows execution is claimed. See [shared context and helper follow-up](checkpoint/dynamic-context.md). 2026-11-06 physical-registration continuation:12 copied-code registration/loader cases pass with complete emitted callback/order/count/flag/transform checks. All72 VR04 records/13 class interfaces are inventoried, with a bounded catalogue of21 helper creation/registration edges and12 verified helper RTTI identities. These are source/component results, not a complete observed physical list or whole-original/native Windows gameplay.342 Python checks and the existing28-case managed pair regression pass;539 originals remain unchanged. Runtime code and 924ad2f packages are unchanged. See [registration](checkpoint/physical-registration.md) and [helpers](checkpoint/vr04-generated-bodies.md). Helper retained lifetime, activation and dynamic shared reclip state remain open before live admission. 2026-11-06 pair lifecycle checkpoint:28 copied-original pair/sphere/exclusion/ base-contact cases match the C# dispatch component's complete callbacks, contact masks, integration/enabled flags and roots. Fixed reference tests also pass without local traces. Actors, response hooks and box results are synthetic adapters; mode2 geometry, whole original and native Windows behavior are not established. The component is not admitted to live gameplay. All338 Python tests, targeted dynamic/body-ownership/Director checks and Debug build pass; Windows cross-publish succeeds with a NU1900 vulnerability-metadata warning.539 originals are unchanged. See [pair contract](checkpoint/pair-lifecycle.md). Both packages identify clean 924ad2f, with519 exact original copies,4285 matching scene files and7 matching prepared audio files each. Standalone Linux from /tmp without SDK/development library paths passes the Carl control replay with4 readbacks identical to ea04210. That run disables validation; no new enabled-validation claim is made. The current Windows package is cross-published/resource-verified; the earlier Wine64 evidence belongs to ea04210 and native Windows testing remains pending. 2026-11-06 TSK startup correction: the original menu creates backgroundLevel1D/ Jimmy before successful TSK story entry, then retains that player ahead of the new level-game input callback. Program now models this on first story entry as well as retained transitions; direct fresh-scene order stays separate. The baseline native Vulkan replay fails on first neutral tick0 with destination roll−10.000001. Corrected first input and300 idle ticks stay upright, preserving heading, held-input return and body/fire continuation with zero Vulkan errors. Eight managed Carl trajectories coverLeft/Right, natural/skip and repeated scenes. All336 Python+2 HUD, managed projects and36 GPU/render passes succeed. Native Linux SDL/Vulkan1920×1080 control replay and VR04 powered keyboard/steering/resize pass with zero validation errors. The returned-frame readback is inspected upright. Final build has zero warnings/errors; Windows x64 cross-publish succeeds. Both packages identify clean ea04210 and have519 matching original copies,4285 scene files and7 prepared audio files each. Standalone Linux from /tmp without SDK/ development-library paths passes the control replay with4 identical development readbacks. Its validation is disabled; development validation is enabled with zero errors. Removing the optional validation layer's library path produces ErrorLayerNotPresent; that diagnostic is preserved separately. The Windows package passes the32-frame Carl replay under isolated Wine64, including first neutral input/300 idle ticks and held-input return. This is preliminary Wine evidence, not native Windows verification. All539 originals are unchanged. Source details: [menu/player lifetime](gameplay/main-menu-player-lifetime.md). Normal landing bank and the actual movement/facing algorithms are preserved. Airborne control return, whole-original timing/rendering and native Windows comparison remain separate. The50 new copied-original derived-caller cases and static body/visual-sync audit are bounded source/component evidence. Matched Wine startup diagnostics progress past MAIN only after an explicit OS repaint acknowledgement; they show background allocation with black frames and perturbed timing, not a playable Carl reference. Earlier checkpoint (2026-11-05): post-skip tilt was still open. User original comparison specifically identifies skipping Carl; the previous validation covered entry and held movement on resume, not subsequent idle. A48-case managed port diagnostic isolates stale mode/turn republishing a bank target in fresh order. New original skip/control/ registration evidence is documented, with its causal gap explicit. Gameplay changes are paused; packages remain dae13ac. See [focused decompilation](gameplay/jimmy-cutscene-decompilation.md). Rotation handoff feedback (2026-11-05): CAM/MCA clear both live and destination pitch/bank while preserving current yaw/dirty.20 isolated original cases compare; managed real banked-jump handoffs and native Linux control restoration pass. Seven source grounded-input/facing trajectories reproduce residual idle tilt, so no generic landing snap was introduced. Whole-original frame/native Windows comparison remains open. See [handoff](gameplay/jimmy-rotation-handoff.md). Rotation packages identify clean dae13ac;519 original copies,4285 scene resources and seven prepared audio files match per platform. Standalone Linux handoff passes with three identical development readbacks and zero Vulkan validation errors. All336 Python/targeted managed checks pass;539 originals unchanged. Windows is cross-published/resource-verified, not native runtime-verified. Turning/fire feedback (2026-11-05): source left/right turn clips restored for low-speed grounded turning; jetpack fire now shares full owner pitch/yaw/roll and visual alignment. Player visual/movement/data/director/jetpack/Python checks pass, native Linux flight and both clip previews render with zero Vulkan errors. Full source morph timing/native Windows comparison remain open. Details: [turn/fire](gameplay/jimmy-turn-fire-fixes.md). Turn/fire packages: clean f0be850,519 original copies and4285 prepared files each. Standalone Linux powered flight and turn previews pass with five identical readbacks and zero validation errors;539 originals unchanged. Windows has build/ resource verification only. Runtime identity is retained in package BUILD-INFO. Body contact foundation (2026-11-05): finite XYZW publication, independent body modes, static-context retention and landing callback ownership implemented.100 full native numeric outputs and six state/branch comparisons pass; existing 20 body/13 Scooter/13 Grappler groups pass. Native Linux both Scooter scenes, Jimmy walking/flight and control phases pass with zero Vulkan errors; all five Jimmy final/airborne readbacks match the preserved air bridge exactly. No special Jimmy contact or broader scene admission is enabled. Evidence and remaining limits: [body contact](checkpoint/body-contact-implementation.md). Body foundation final validation: full scripts/test.sh exits0 (333 Python+2 HUD, 45 managed projects,36 main Vulkan completions, asset/save/audio checks and539 original hashes). Clean c9d0c5b Linux/Windows packages have matching519 original copies and4271 prepared files each. Standalone Linux from /tmp passes both Scooter scenes and walking/jetpack, with27 matching development readbacks and zero validation errors. Original/native Windows comparison and actual special Jimmy callbacks remain separate; see the body implementation report. Jimmy airborne bridge (2026-11-05): ordinary forward/back pitch and turn lean, source target/dirty/Init/retained-seed ownership, full authored Euler and logical root hierarchical rendering are integrated.146 native component outputs match; final OMT model decomposition remains an explicit precision gate, and sole alignment remains a modern local adapter. Native Linux all-five-jump walking/ jetpack replays pass with25 identical cross-binding captures; Carl/control phases pass, zero validation errors. Collision admission is unchanged. Detailed current validation and limits: [air bridge](gameplay/jimmy-air-bridge.md). Air-bridge package validation: clean27be8ed for Linux/Windows;519 original asset copies and4271 prepared files match per package. Standalone Linux from /tmp passes walking/jetpack, all five final/airborne captures matching development, zero validation errors. All script checks pass across recorded runs:325 Python+2 HUD, 44 managed projects,36 main render completions and539 unchanged original hashes. Native Windows/full original comparison and exact final OMT decomposition remain open; the current bridge preserves the explicit modern sole alignment. Jimmy movement (2026-11-05):108 copied-original ground/air/facing/jump observations match managed components.26 integrated movement/20 body groups pass; lab and two-way underground routes retain the user's baseline. Native Linux walking and VR04 powered input replays pass all five jump aliases and resize, with identical readbacks across aliases and zero Vulkan validation errors. Carl control lock, control-independent body/fire and Grappler native replays pass. Source pressure is restricted to real source support; the lab/outside capsule keeps compatibility behavior. Ordinary air pitch/full root rendering and special/dynamic contacts remain open, as do complete original and native Windows comparison. Details and provenance: [Jimmy stabilization](gameplay/jimmy-movement-stabilization.md). Jimmy final package validation: both self-contained builds identify cleanacd1b28; 519 original copies and4271 prepared resource files match per package. Standalone Linux from /tmp repeats walking/powered SDL replays with two matching development readbacks and zero validation errors. All script checks pass across documented run/resumptions (320 Python+2 HUD,43 managed projects,36 main render completions, asset/audio/save checks,539 unchanged originals). Native Windows and whole-original movement comparison remain pending; see the stabilization report for exact gates. Scooter feedback follow-up (2026-11-04): missing Goddard follow pitch/roll leveling and source positional engine-frequency correction are implemented.13 Scooter/2 ScooterAudio/10 Patrol groups,24 targeted Python checks, route/audio regressions and both10-phase native Linux SDL replays pass; zero validation errors.11 new native facing traces and11 frequency traces distinguish source math from port composition. The correction lowers engine pitch slightly; audible parity/native Windows confirmation remains unverified. Reports: gameplay/scooter-pitch.md and scooter-dismount-leveling.md. Both packages identify cleanfccc0b9 with14 matching Scooter resources. Standalone Linux from /tmp passes both10-phase SDL replays;20 readbacks match development byte-for-byte.539 originals remain unchanged. Provenance: scooter-live.md. Scooter (2026-11-04): live in VR05 and outside LEV1 through normal owned gadget selection and F1 grant.53 copied-original component observations match managed math; actual-level tests cover mount, wheel/body phases, drive/jump, head-on recovery, camera and cleanup/RETURN. Native Linux GTX1070/private Xvfb1920×1080 passes both10-phase SDL input replays with zero enabled Vulkan validation errors. A2560×1080 Player-page capture shows all32 rows including Scooter; inspected. Headless replay passes both scenes. Source and final validation/package provenance are in gameplay/scooter-live.md. Fixed cadence, Jimmy's outside capsule, provisional NPC/dynamic door body, other levels and original/native Windows comparison remain explicit limits. Both packages identify clean62c07a3; all14 Scooter resources match. Standalone Linux from /tmp passes both10-phase native SDL replays, with all20 stage captures identical to development.303 Python+2 HUD/all managed/36 main GPU regressions and the post-package539-file audit pass. Historical gadget-only status below is retained chronologically. Compilation, native runtime, original behavior and full feature completion are separate evidence categories. **The game is not fully reconstructed/playable yet.** Current Grappler scope is liveVR08: source ordinary body/shared sphere context, recovered camera, original rope/SWING/cursor, sounds and selection/RETURN cleanup. 113 original-component observations;13 managed live/40 Python checks pass. Native Linux GTX1070 and headless replay pass9 phases with zero Vulkan validation errors. Both packages identify clean43806fe and contain9 exact Grappler resource files. Standalone Linux from /tmp passes and matches all9 development captures. Full scripts/test.sh passes297 Python+2 HUD/all managed and route/gadget/audio checks; all539 original hashes verify unchanged. Exact art/capture and package evidence: gameplay/grappler-live.md. Whole-original VR08, native Windows, other levels and general dynamic pairs remain unverified. Historical component-only status (superseded for boundedVR08): Grappler query/constraint checkpoint:42 new copied-original observations (72 total),10 managed/18 Python groups,11 camera/13 Rocketship groups and main build pass. ActualVR08 ray/filter/material integration works in tests. Lab gameplay readback is byte-identical; lab and VR08 geometry preview render with enabled Vulkan validation, zero errors.539 original hashes verify. Attached world-body/ base-contact composition, camera and helper presentation/lifetime remain open; Grappler is not live and playable packages remain70cba65. See gameplay/grappler-query.md. Grappler scalar checkpoint:30 copied-original Wine32 observations match5 managed groups,14 Python source/projection checks,25 movement/15 gadget groups and main build pass. No live Grappler, new Vulkan path or original/native Windows gameplay comparison is claimed. See gameplay/grappler-component.md. Playable packages remain70cba65 with the orbit horizontal-input correction, verified by native Linux SDL input/resize and zero enabled Vulkan validation errors. Orbit inspector follow-up: corrected right-handed inspection view/perspective to gameplay's positive-Z left-handed convention.36 camera/aspect cases and all existing projection checks pass. Enabled-validation llvmpipe captures confirm the removed horizontal reflection; lab gameplay captures are byte-identical. Native/package evidence: orbit-camera.md. Native Windows remains unverified. Native GTX1070/private Xvfb orbit input/resize check passes, enabled validation and zero errors. Both packages identify clean d8f1b51. Standalone Linux capture from /tmp matches the corrected source byte for byte;539 original hashes verify. Rocketship impact follow-up: corrected first-material red-flash RGBA previously misapplied to ship scale.13 managed groups include actual wall contact plus130 frames of unchanged geometry, correct ship-only tint, recovery and repeated renders. Source guards independently verify material allocation and DLL diffuse consumption; old numeric probe labels were insufficient semantic evidence. See gameplay/rocketship-impact-render.md for REA and package/runtime checks. Both packages identify clean62e908e. Standalone Linux from /tmp passes all10 ordinary flight/control phases on GTX1070 with real SDL keys and enabled Vulkan validation, zero errors.22 targeted Python checks and539 original hashes pass; native Windows runtime for this fix remains pending. VR02 Rocketship (2026-11-03): live aircraft/static support and callbacks, original input/facing/maneuvers/camera, owned rider/Goddard/fire and engine lifecycle. 14 flight/12 final integration groups;96 isolated original component observations. Native Linux GTX1070/private Xvfb1920×1080 ten-phase real SDL replay passes with zero Vulkan validation errors; headless matches.295 extra ordinary ticks are an accelerated port fixture, not original cadence or manual tutorial completion. Full268 Python+2 HUD/managed/33 GPU+17 material regression passes;539 hashes verify. Both packages identify clean e77add5;18 resource files/four source RGBA hashes match. Final standalone Linux from /tmp passes all10 SDL phases with validation enabled. Packaged1920×1080 F1 Player page fits32 lines (headless layer unavailable). Other levels, special missions/hazards/RNG, full source pair discovery, generic Jimmy post-vehicle Euler/render bridge and original/native Windows comparison remain boundaries. See gameplay/rocketship-live.md for package provenance. Give All Gadgets debug action (2026-11-03): F1→Player grants items0..7 without changing selection. Debug-menu action/navigation/repeat/layout checks and main build pass; llvmpipe headless1920×1080 view renders all rows with zero Vulkan validation errors. Readback inspected; artifacts/debug-give-all/render.log. Rocketship component/Grappler recovery (2026-11-03): isolated keyboard/vertical math matches28 copied-original Wine32 observations across13 managed groups. Rocket source3/resource5/Grappler12 and existing Gadget15/Movement25 groups pass; main .NET10 build succeeds with the existing NU1900 metadata warning.539 originals verify unchanged. Exact resource export remains opt-in; no live vehicle/Grappler, new Vulkan/native Windows or whole-original gameplay verification is claimed. No package rebuild: playable d6ca417 remains baseline. Next live gates and full scope are recorded in gameplay/rocketship-component.md. Supplied jn-engine assessment (2026-11-03): documentation/source/artifact review only; no new build or runtime validation. Same declared EXE, different declared OMT2.dll;254 ASE/126 PNG/35 GAM files match byte for byte. All539 original hashes verify. Class specs, decompiler bodies and derived D3D7 fixtures are useful evidence candidates; proposed original gameplay traces/raw OMTC/Ghidra project are not included. No prototype C mechanics were imported or port fidelity promoted. See reviews/jn-engine-survey.md; existing package claims below stand. Baseball (2026-11-02): source hold/release/20-slot pool, exact ball/aim/TIME art, bounded ball body, VR07 moving-target callbacks and F1 grant are implemented. 12 managed/6 visual/16 Baseball Python groups and independent review pass. Native Linux GTX1070/private Xvfb1920×1080 passes real SDL Ctrl/Left/Up/Tab, all eight render phases, actual target hit/respawn and zero enabled Vulkan validation errors. Its1010 extra ordinary player/target ticks are an accelerated port fixture. Sphere contact discovery, angular wall response, full enemy combat, retained helper lifetime and original/native Windows comparisons remain open. Complete regression passes232 Python+2 HUD, all managed suites,32 GPU runs and 17 material checks;539 originals verify. Both packages identify clean d6ca417; all16 Baseball/aim pixel hashes match. Standalone Linux from /tmp passes all eight phases and the1920×1080 F1 Player view; release validation layers unavailable. All539 originals verify after packaging. Native Windows remains unverified. gameplay/baseball-implementation.md records the coverage map. Scooter's recovered wheel/control dependencies are documented but not implemented as invented Jimmy movement. Bubble (2026-11-02): original mode2 controller, sprite/flags/pulse/fade, controlled drain, loop/boing ownership, cleanup650 and boundedVR01 source body/floor111 are integrated. Six isolated copied-original body references match managed output. Independent review and13 integration/11 motion/8 visual groups pass. Full suite: 202 Python+2 HUD, all managed checks,31 main GPU completions and17 Vulkan material checks, zero enabled validation errors; SDL dummy lifecycle and539 originals pass. Native Linux GTX1070/private Xvfb1920×1080 passes real Ctrl/Left/Shift/Tab and all eight rendered phases. The fixture advances306 extra normal player ticks; this is not original cadence/manual completion/full hazard or original visual parity. Lab/outside capsule, fixed render bridge, generic camera/Euler scheduling, other damage flags, mouse/analog and retained-helper lifetime remain explicit limits. Both self-contained packages identify clean1ae5f42 (source_dirty=no), preserve exact Canv1 RGBA hashes and publish successfully despite NU1900 metadata warnings. Standalone Linux from /tmp passes all eight Bubble phases and a1920×1080 F1 Player view; release validation layers are unavailable, distinct from enabled checks. All539 originals verify after packaging. Native Windows remains unverified. See gameplay/bubble-implementation.md and artifacts/bubble-implementation/. Invisible continuation (2026-11-02): recovered mode2 timers, fixed energy cost, alpha/blend flashing, original meter and companion/cutscene cleanup are integrated. Shared source energy warning latch/audio now serves ray/jetpack/Invisible. Fifteen managed groups and182 Python source/parser checks pass. Native Linux/private Xvfb GTX1070 at1920×1080 verifies real SDL Ctrl/Tab and six rendered phases with zero enabled Vulkan validation errors; headless matches. The fixture accelerates 836 ordinary player ticks between captures and does not prove original cadence, enemy immunity, original pixel placement/visual parity or tutorial completion. Full scripts/test.sh passes182 Python+2 HUD, all managed suites,30 main GPU completions and17 isolated Vulkan material checks, with zero enabled validation errors. SDL dummy lifecycle and539 original hashes pass. Both self-contained packages identify clean1897dd2 (source_dirty=no) and preserve exact Canv21/22 meter pixel hashes. Standalone Linux from /tmp passes all six Invisible phases and a1920×1080 F1 Player-page render. Release validation layers are unavailable, distinct from enabled source/native checks. Both platform publications succeed despite NU1900 vulnerability-metadata warnings. Windows native gadget verification remains open;539 original hashes verify again after packaging. See gameplay/invisible-implementation.md and artifacts/gadget-followup/. Gadgets (2026-10-27 UTC): original gadget screen and combined shrink-ray/jetpack are implemented; actual VR04 soldiers use original shrink morphs, source timer, authored AI reset and ordinary10-point score/save effect. First complete run: 157 Python+2 HUD, all managed suites including15 gadget groups,29 main GPU completions and17 isolated blend/readback checks, zero enabled validation errors. Heading matches15 original-DLL references and65,540 sweep cases. Final ordering/ authored-mode fixes pass the complete rerun in integration-final.log. Native Linux GTX1070/private Xvfb verifies1920×1080 menu input/pause/capture and 1280×720 held ray against an actual authored VR04 target, with zero validation errors. The fixture moves only the player near a settled soldier; it does not establish gameplay navigation to the platform or original visual parity. Other effects, full enemy AI/combat, energy warnings and lifecycle gaps remain; see gameplay/gadgets-implementation.md. Both self-contained packages identify clean7b5230c (source_dirty=no), with54 canvases and all new animation manifests. Standalone Linux from /tmp passes widescreen menu and90-frame ray fixtures; release validation layers are unavailable, distinct from enabled source/native checks. Windows publishes successfully; native Windows gadget testing is not claimed. NU1900 vulnerability-metadata warnings do not prevent publication. All539 originals verify again after packaging. Logs: artifacts/gadget-implementation/. C4 ordinary body (2026-10-27 UTC): all ten original-code body references match managed composition. VR04 uses source static response/support/friction and live radius60/100; lab/outside retain the capsule. Twenty body groups plus full suite pass:129 Python+2 HUD, managed checks,27 zero-error enabled-validation GPU renders, SDL dummy lifecycle and539 unchanged originals. Native GTX1070/private Xvfb at 1920x1080 passes300-frame jetpack input/resize/steering/landing. Logs/readback: artifacts/body-composition/. Source scope/independent review are linked from checkpoint/collision-body-implementation.md. Original dynamic actor eligibility, special callbacks, logical/render pivot and modern noclip occupancy remain open. Native Windows and whole-original-game comparisons remain unverified. The final expanded Python suite passes130 checks after the Init authored/live speed closure; the129-check full-integration log predates that extra regression. Both self-contained packages identify clean35a2a66. Standalone Linux from /tmp passes source-body tutorial flight and lab/tutorial/RETURN assertions. Release validation unavailable; enabled source/native checks separate. Both publish steps succeed despite NU1900 vulnerability-metadata warnings. Windows remains cross-built; 539 original hashes reverified after packaging. Package logs: artifacts/body-composition/. C4 static composition follow-up: seven copied-original dispatcher fixtures match the isolated managed grid/retry component; ten copied479D20/support fixtures complete as references for future body composition. No live capsule replacement is claimed. Prepared scenes retain signed material IDs; actual lab/outside special contacts prevent ordinary-only assumptions. See checkpoint/collision-body-probe.md. The temporary automatic approval usage-limit failure was resolved through normal review after its stated reset; original component execution then completed. Full scripts/test.sh passes122 Python plus2 HUD checks, managed suites and27 zero-error enabled-validation GPU completions;11 focused dispatcher groups pass. SDL dummy lifecycle and539 original hashes pass. Evidence is under artifacts/collision-integration/. No new native Windows or whole-game original comparison; preserved baseline and prior release packages are unchanged. Linux-x64 and win-x64 self-contained publish succeeds into separate build artifact directories. Restore's NU1900 records unavailable vulnerability metadata; compile/ publish succeeds. No new distribution package or native Windows execution claimed. C4 continuation (2026-10-26): REA/bytes recover the bounded47BD80 sphere kernel; ten isolated original Wine32 synthetic cases pass with explicitly guarded fixture rebasing. The managed component matches numeric references; live capsule replacement remains gated. Source-root overhead velocity reflection/mode1 jump/fire cancellation is integrated for LV1B/VR04/LEV1, including without control and with the modern camera query disabled. Full sound/gadget cleanup and other scene filters remain. Registration/repeated Init and shared-eye handoff reports distinguish recovery from pending runtime changes. See checkpoint/collision-response.md and linked reports. Final scripts/test.sh passes in artifacts/collision-response/integration-final.log: 104 Python checks plus separate HUD checks,25 movement groups, five new sphere groups, remaining managed checks,27 zero-error enabled-validation GPU completions, SDL dummy lifecycle and539 unchanged original hashes. Both input-only routes use the source geometry queries and retain their progression/contact checks. Native Linux GTX1070 at1280x720 passes Carl/control, jetpack SDL keys/resize and debug tools/checkpoint reload (four commands, five renders), zero validation errors. The original component probe is not whole-game or native Windows validation. Both self-contained packages identify clean57dfcc6, source_dirty=no. Standalone Linux from /tmp passes the control-phase fixture and nondefault tutorial/RETURN assertions. These release invocations lack validation layers; enabled source and native-window evidence above is separate. Logs: artifacts/collision-response/ package.log, package-control.log and package-return.log. Baseline untouched; Windows cross-publication adds no native Windows runtime claim. Control-phase continuation (2026-10-26): body/fire eligibility and ordered raw preparations are source-byte guarded separately from port behavior. Targeted24 movement groups, coupled director/preparation tests and the airborne control fixture pass. Native Linux GTX1070 at1280x720 passes Carl/control-phase, F1 pause/ Force Control and F3/tools/same-window checkpoint checks, all with enabled Vulkan validation and zero errors. The input-only route uses the actual binding and Down braking at turns; source score/task/inventory/contact checks still pass. Detailed gaps: checkpoint/control-phase-implementation.md. Native Windows and original whole-game comparisons remain unverified for this chunk. Full scripts/test.sh passes in artifacts/uncontrolled-phases/integration-final.log: 93 Python checks,24 movement groups plus remaining managed groups,27 zero-error GPU renders with enabled validation, SDL dummy checks and539 unchanged hashes. REA6.8.0/Ghidra12.1.4 decompiled five selected procedures locally using a generated copy; complete Evidence is retained.421FB0 lacked a recognized procedure; that limit is recorded without replacing its existing byte/vtable evidence. Static pseudocode is not original execution. See rea.md for local installation and CLI. Both self-contained packages identify clean commit c285b57 (source_dirty=no). Standalone Linux launches from /tmp pass the airborne control-phase fixture and nondefault tutorial/RETURN scalar, root, companion and availability assertions. Release validation layers are unavailable; enabled-validation results above are separate. Logs: artifacts/uncontrolled-phases/package.log, package-control.log and package-return.log. Preserved baseline untouched; Windows cross-publication adds no native Windows runtime evidence. Retained-state continuation (2026-10-26):14 new source-evidence tests and22 movement groups cover named property loading, cleanup/Init/PostLoad/control boundaries and bounded route scalar carry. Corrected full tutorial/lab gadget availability and TSK/unknown-state tests pass. An isolated Wine32 component probe executes copied original vector accessors and43A750 with synthetic fixture vtables and invalid sound handle; source targets are byte-guarded. It confirms body-field writes and unrelated sentinel preservation, not full loader/gameplay or native Windows behavior. Evidence: artifacts/retained-lifecycle/original-probe/wine.log. Integration:84 Python checks, all managed groups and26 GPU render checks pass with zero enabled-validation errors. Final nondefault mode1/hold.625/fire6.125 ordinary-load replay independently checks jump/body resets on tutorial and RETURN; all539 original hashes unchanged. A fixed200ms effects-smoke baseline failed under load; paused stream construction plus a bounded consumption poll removes that sampling assumption (the initial failure did not record its queue baseline). All four SDL dummy audio checks pass in audio-final.log; final route is in route-final.log. No new native Windows runtime verification. Both self-contained packages rebuilt from clean9ad5a8e, source_dirty=no. Linux standalone launch from /tmp passes nondefault lab→VR04→lab scalar/body/RETURN assertions, with the same presentation instance retained. That release invocation has validation unavailable; source GPU enabled-validation evidence remains separate. Logs: artifacts/retained-lifecycle/package.log and package-route.log. Debug expansion (2026-10-25): full scripts/test.sh passes, with70 Python checks, 21 movement groups, debug page/overlay groups, real SDL dummy pause/mute timing, remaining managed/GPU routes and539 unchanged original hashes. Native Linux GTX1070 at1920x1080 passes F1/Tab/F3, flight during a source cutscene, input capture, camera/audio controls, checkpoint save/load and same-window rebuild. Enabled Vulkan validation reports zero errors. Headless Player1920x1080 and Systems2560x1080 screenshots fit; see artifacts/debug-expansion/. SDL_SetAudioStreamGain is present in the packaged Windows SDL3 DLL export table. Cross-build/export evidence does not add native Windows runtime or original-game behavior verification. Both platform packages rebuilt from clean d783898, source_dirty=no. Standalone Linux package launched from /tmp with Player page/noclip1920x1080 passes on llvmpipe; that invocation lacks validation layers. Evidence: package.log and package-smoke.log under artifacts/debug-expansion/. Latest implementation continuation: controlled player phase ordering, route static camera query and PostLoad gadget availability pass the full `scripts/test.sh`. Evidence: artifacts/implementation/integration-final.log;70 Python regressions, 20 movement groups,11 new camera groups,13 existing support groups, remaining managed groups, GPU story/practice/RETURN/Carl/objective/jetpack/animation checks, SDL dummy lifecycle and539 unchanged original hashes. GPU validation is enabled and every run reports zero errors. Native Windows runtime and original-game comparison are not added by these tests. Known limits remain explicit in the current checkpoint coverage/implementation reports. Both self-contained packages now identify clean09cf117. Linux release smoke from `/tmp` passes practice/RETURN (including input order and gadget availability) and outside load; see linux-package-return.log and linux-package-outside.log under artifacts/implementation/. Those release invocations have validation unavailable; the enabled-validation evidence is the full integration run above. Windows is cross-published, with no additional native Windows runtime claim. Architecture checkpoint: see `checkpoint/review.md` for coverage and remaining gates. C1 objective/task order and C2 per-actor Carl callback corrections pass the full `scripts/test.sh` integration, all GPU runs with zero Vulkan validation errors. Final Python discovery passes62 tests including the additional source evidence follow-ups. All539 original hashes remain unchanged. These checks do not add native Windows or original whole-game runtime verification; earlier Windows user reports remain limited to their tested builds and route. Self-contained Linux/Windows packages rebuilt from clean531883a. The Linux release package passes the60-frame Carl held-input/control-lock smoke from `/tmp`; validation layers were unavailable for that invocation. Current Windows release is cross-published, not newly native-runtime verified. Later loader-provenance edits affect evidence/tests only; runtime code is unchanged from that package. | Feature | Implemented / tested | Evidence and limits | |---|---|---| | Original preservation | Verified after extraction/runtime tests | All539 original SHA-256 hashes unchanged | | GAM Python / C# | All35 files,3299 entities | Exact Python round-trip; C# corpus/start/corruption checks | | OMT recovery | All100 archives audited | 2155meshes,2183materials,1614textures,875RIFF, zero failures | | C# original loaders | Linux runtime | All128 PNG /254 ASE base meshes; morph/base UV tests | | .NET10 Linux build | Pass, zero warnings/errors | Local SDK10.0.402/runtime10.0.13 | | Windows x64 publish | Pass | Correct PE32+ app and SDL3.dll; initial native rendering user-reported | | Windows story/practice package under Wine64 | Preliminary headless/window pass |12s story;12-frame practice pause/N/resume/resize and original streams; simulation matches Linux; validation disabled | | Corrected Windows preview under Wine64 | Preliminary window pass | Six frames, key2 animation switch, resize800x600, exit0; validation disabled | | 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 | 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 | | Null-material visibility | Binary rule implemented/tested | Original DX renderer skips null pointers; proxy geometry retained | | SDL3 presentation/input | Linux Xvfb + GTX1070 | Six frames; injected keyboard event received | | Swapchain resize | Linux tested | 960x720 ->800x600, zero validation errors | | Persistent level presentation | Native GTX1070 six-load pass |One window/renderer; stable IDs/core resources, bounds/focus/fullscreen; invalid load retains rendering; zero cumulative Vulkan errors | | Widescreen HUD/debug UI | Managed matrix + native captures/pointer pass |Uniform proportions, centered modal/F1, anchored HUD/F2;4:3/16:9/21:9/32:9, DPI and strict boundary checks | | F1 debug menu | Managed/SDL audio tests + native Linux GTX1070 window pass |Pause/actions/input capture/resume/800x600 resize, zero validation errors;33 prepared scene/start choices; intentional modern tools, no new Windows runtime claim | | F1 Force Control | Menu/director/interaction checks + native Linux SDL pass |Persistent override through source shots; disable/reset restores normal ownership; actual movement with neutron1a active, zero Vulkan errors; music handoff regression passes | | F2 live debug overlay | Managed timing/layout/menu + native Linux SDL pass |Non-pausing movement, visibility/repeat/menu priority and800x600 resize; all checks true, zero Vulkan errors; session persistence; modern tools, no new Windows runtime claim | | Focus / fullscreen | Code present, partial/unverified | Focus clears keys; fullscreen toggle not dynamically tested | | 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 |18 movement groups + GPU replay pass |Corrected mode1 fall-through to source air handler: acceleration/braking/turning during jumps and bursts; RETURN root placement; phase/collision/lean remain provisional | | Static source collision |11 deterministic groups + source NC byte checks pass |2851 lab collision triangles after excluding18 NC-shape triangles; all2869 render/source triangles and325 null-material proxies retained; solver/dimensions provisional | | Simulation cadence |18 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 | 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/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 | | RETURN / scene reload |Source/director tests + Linux headless pass |Lab/VR04/lab restores departure root/companion using one renderer, retains STRT yaw and suppresses intro; fresh TSK loading separate; original fades unverified | | Supplied scene availability |All35 GAMs supplied;33 have OMT geometry |Level3B/E GAMs are duplicate four-entity records apart from level code; registered OMT archives absent, normal gameplay use unverified; no substitution or established blocker | | Lab task/contact route | Managed input replay + headless GPU pass |887 input-only movement steps with live contacts/NPC updates reachSCENE60, score150, jetpack andlevel1; separate direct-contact GPU rebuild check | | Lab underground passage |2763-step managed input replay pass |Both switches,2 clear crossings and timed reclosure; closed-gate negative sweep; explicit riverlab/SCENE65 fixture, prior progression and original comparison unverified | | 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 | 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 | 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, resize-input.ppm, package-linux.ppm, extracted/audit.json and reference*/ logs. The lab/player images have been visually inspected; they do not establish original 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. Subsequent RETURN/underground checkpoint: artifacts/return-underground-tests.log passes the complete suite (41 Python checks,18 movement groups, source checkpoint checks, actual lab/VR04/lab rendering, underground input replay and539 unchanged originals). RETURN restores the exact departure root and companion, retains STRT yaw and suppresses intro, with zero Vulkan validation errors. NC registration exclusion fixes the underground water-plane rise while retaining rendering. Both Linux x64 and Windows x64 self-contained executables republish successfully. No subsequent native Windows/Wine runtime check is claimed. F1 menu checkpoint: artifacts/debug-menu-full-tests.log passes the full suite, including actual SDL audio pause/resume and539 unchanged originals. Native Linux window/input/action/resize evidence is artifacts/debug-window.log; screenshots were visually inspected. Both Linux and Windows self-contained packages publish successfully. No new Windows/Wine runtime check was run for this addition. Live F2 diagnostics: artifacts/live-debug-full-tests.log passes the full suite and539 original hashes. Native SDL live movement/menu/visibility/resize and the existing force-control check pass with zero Vulkan errors. Headless lab/VR04/lab retains visibility and samples1/2/3 with zero errors; camera/layout/timing groups pass. See artifacts/live-debug-window.log and live-debug-return.log. No new Windows runtime claim is made. Both self-contained platform packages rebuild successfully with this addition. Persistent presentation/widescreen checkpoint: full regression suite passes in artifacts/persistent-widescreen-full-tests.log, including539 preserved hashes. Native6-load/one-window/one-renderer fullscreen/bounds/focus/pointer checks pass with zero enabled-layer errors (artifacts/window-transition-final.log). F1/force and F2 input checks still pass. Both packages rebuild; packaged Linux1920x1080 lab/VR04/lab retains one renderer and exact RETURN state with validation unavailable. No new native Windows runtime 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, and no original comparison captures. Walk evidence: artifacts/walk-tests.log, test-walk.ppm, walk-start.png and walk-window.ppm. Original fidelity is unverified.