# Progress — 2026-11-06 Workflow update: the user authorizes pushing completed, verified commits to the existing `Gitea/main` upstream from now on. See docs/decisions.md and AGENTS.md. Next-work plan: [recover interactions, then implement verified changes](checkpoint/next-work-plan.md). It defines the Carl source/differential gate, the bounded implementation step, movement/contact/animation recovery and the end-to-end platform checkpoint. ## Carl neutral return: TSK startup lifetime correction [Menu/player lifetime](gameplay/main-menu-player-lifetime.md) closes the missing ordering distinction: the original initializes a backgroundLevel1D/Jimmy before a successful TSK story load. The loader retains Jimmy and appends the new level game later. First playable story entry therefore uses player-before-input, just like subsequent retained transitions. Fresh direct-GAM order remains separate. Program previously used that order only when carriedSave was non-null. The correction includes TSK-backed story entry and adds no movement/Euler reset. The strengthened Vulkan control-phase replay fails on baseline neutral tick0 (destination roll−10.000001), then passes first input plus300 idle ticks after the correction, preserving heading and held-input/body/fire continuation. Eight managed 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 first returned-frame readback is inspected upright. Final build has zero warnings/errors; Windows x64 cross-publish passes. Source/ test provenance is in artifacts/main-menu-player-lifetime/verified-change.json. Clean package validation follows; historical a2e89c6/dae13ac is not the identity of an in-progress cross-build. Native Windows/original Carl execution stays unverified. The additional derived-frame caller50 native cases and static body/visual-sync audit close their bounded omissions. A matched original-resolution Wine startup diagnostic progresses to backgroundJimmy allocation only after the recorded repaint acknowledgement; frames remain black. See [startup observations](gameplay/original-startup-diagnostic.md). These are perturbed allocation observations, not full original Carl or Windows fidelity. ## Earlier Carl lifecycle investigation and original-state observer The accepted plan is underway; its gameplay correction gate remains open. [Integrated lifecycle trace](gameplay/jimmy-cutscene-lifecycle.md) records292 new bounded original-code cases:96 contact,8 input/controller/skip-prefix,48 task/ objective and140 base/STOP. Complete DEFAULT pseudocode is recovered through a verified mapped analysis carrier. None supplies a source-supported movement reset; the original components reproduce the port's order-sensitive stale-bank writer. A read-only Windows32 observer now launches or attaches to a verified **copied** original. Wine32 attach/read/leave-running integration passes928 samples, with observer exit0 and original still running. Local original rendering remains black: only MAIN, no Jimmy/gameplay reference, was observed. Native Windows Carl state, actual entry mode/turn, registry order and first-resume timing remain required. All539 originals are unchanged. Runtime a2e89c6/packages dae13ac and the48-case port diagnostic remain unchanged; no new gameplay fix or fidelity claim is made. ## Decompilation first: Carl skip-to-idle remains open User reports the remaining tilt after skipping Carl and confirms original landing bank is normal. Further gameplay changes are paused for a focused decompilation checkpoint. Runtime a2e89c6/package dae13ac is preserved. The earlier correction covers entry/locked state; its moved-input resume check missed neutral return. A read-only48-case port trace isolates the first stale target write: all24 fresh input-before-player cases retilt, all24 retained-order cases stay upright. Original control/default/cleanup boundaries do not prove a reset; source skip writes a Jimmy action latch plus shared skip state absent from the current fast-forward model. Three new REA/Ghidra pseudocode results and guarded actual GAM/contact/ registry findings are retained. No gameplay fix is inferred from these local facts. See [decompilation gate](gameplay/jimmy-cutscene-decompilation.md), [port trace](gameplay/jimmy-cutscene-port-trace.md), and its source reports. Research checkpoint checks: both source guards and all48 recorded diagnostic cases/provenance verify,539 originals unchanged, no src/tests/scripts changes. Source explanation of upright return after original skip is still open. This checkpoint preserves the reproducible failure and narrowed investigation; it does not mark the post-skip bug fixed or authorize speculative gameplay fixes. ## Jump tilt at cutscene and landing handoffs Recovered CAM/MCA direct writes flatten both current and destination Euler to (0,currentYaw,0), preserving dirty; the old port cleared current only, allowing uncontrolled facing to restore the airborne tilt. Fixed that existing boundary, with20 isolated original-code cases, eight integrated handoff cases, and the native banked Carl/control-restoration replay passing. Input, body and fire ownership remain separate; see [handoff](gameplay/jimmy-rotation-handoff.md). The separate idle-after-landing report has seven original ground-input/facing trajectories: neutral input retains residual bank; active input levels normally. No universal grounded snap is supported by those recovered paths, so no landing policy was invented. Whole-original frame ownership remains an explicit question; see [landing evidence](gameplay/jimmy-landing-rotation-source.md). All336 Python tests and targeted managed checks pass; native Linux handoff, Carl and powered-flight replays report zero validation errors. Both packages identify clean dae13ac;519 original copies/4285 scene resources/seven audio files verify each. Standalone Linux repeats the handoff with three matching GPU captures; three flight captures also match the prior runtime.539 originals unchanged. Native Windows/whole-original frame comparison remain open. ## Turning animation and jetpack attachment feedback Restored supplied six-frame jimleft/jimright clips to default preparation and runtime pose/action selection. Standing/slow turns select the source direction; forward running, backpedal, jump and primary action retain precedence. The input side field is retained between its ground-input updates; repeating a pose does not restart its clip. JetpackVisuals now shares Jimmy's full visual transform, including pitch/roll and the modern local alignment, fixing yaw-only fire. New player visual groups,28 movement/14 jetpack, original-data/Director checks, 333 Python tests and source guard pass; build is clean. Native Linux powered replay and both turn previews pass with zero Vulkan errors; captured flame tilt was inspected. Source facts/modern alignment and timing limits: [turn/fire fix](gameplay/jimmy-turn-fire-fixes.md). Body hazard/climb recovery remains the subsequent priority; this feedback fix changes no movement physics. Both packages identify clean runtime f0be850; each verifies519 original copies and4285 prepared resources. Standalone Linux flight/turn checks from /tmp pass with five byte-identical development readbacks and zero Vulkan errors.539 source originals remain unchanged. Windows is cross-published, not runtime-verified. ## Body root/mode ownership checkpoint Implemented the next Jimmy body prerequisites while keeping lab/outside capsule and special-contact gates. OriginalBodyPhysics separates integration/support/ collision/world response/alternate/global geometry, retains previous/current Vec4 and anchor, gives callbacks the live root and uses fresh XYZW equality for candidate publication. Support/landing callback state ownership is explicit; only world publication applies its candidate clamp/falling displacement. New conformance compares100 complete native outputs (19 root+81 mode/landing) and six callback-state cases with an explicit synthetic-vs-real wall boundary. All pass, as do existing20 body/13 Scooter/13 Grappler targeted groups. Source/ provenance: [implementation](checkpoint/body-contact-implementation.md), [root](checkpoint/body-root-publication.md), [modes](checkpoint/body-modes.md). Both investigators returned ownership. Real Jimmy hazard/climb state and dynamic eligibility remain next; no broad admission, new gadget or whole-original parity is claimed. Full scripts/test.sh exits0:333 Python+2 HUD,45 managed projects,36 main render completions, zero validation errors, save/animation/audio and539 original hashes. Both packages identify cleanc9d0c5b with519 original copies/4271 prepared files matching per platform. Standalone Linux from /tmp repeats both Scooter scenes and Jimmy walking/jetpack;27 captures match development exactly, zero validation errors. Native Windows and whole-original comparison remain pending. Final provenance, the nonfatal NU1900 metadata warning and source gates are in the body contact implementation report. ## Jimmy airborne target and render bridge Ordinary jump/jetpack now reads source forward/backward pitch and turn lean, with apply-owned seeds preserved through Bubble transitions and bounded route loads. Destination Euler/dirty, current full STRT/authored Euler and Init are separate. Full Euler is published to the director and character model; the old Bubble-only orientation path is removed. Source degree caching/hierarchical rotation now pivots around the logical root. The upright sole alignment is retained as an explicit modern local correction; final OMT decomposition precision remains open. 146 original observations compare with managed components: prior108,12 ordinary air targets,12 render angle writes and14 hierarchical matrices.14 additional final render matrices document a precision mismatch rather than conformance.28 movement, 20 body and13 Grappler groups pass; lab/two-way underground routes pass. Native Linux1920×1080 walking/jetpack replays pass all five jump bindings;25 final/airborne readbacks match across bindings. Carl/control phases pass, zero validation errors. The existing powered replay now travels359.19 instead of403.88 with source pitch attenuation; rise418.93 and source impulse/gravity remain unchanged. First forward, brake and ordinary jump captures were visually inspected. Evidence/integration: [air bridge](gameplay/jimmy-air-bridge.md), [target state](gameplay/jimmy-air-target-state.md), [render transform](gameplay/jimmy-render-transform.md). Both investigators returned ownership. Remaining priority is root-changing/special/dynamic contact state before replacing the lab/outside capsule; no gadget expansion. All scripts/test.sh checks pass across the recorded initial/corrected Director/ resumed runs:325 Python+2 HUD,44 managed projects and36 main Vulkan completions, zero validation errors, save/load/animation/audio and539 unchanged originals. Both self-contained packages identify clean27be8ed;519 original copies and4271 prepared resource files match per package. Standalone Linux from /tmp with SDK variables unset passes walking/powered replays; all five final/airborne readbacks match development exactly. Windows native and whole-original comparison remain pending. Final provenance and limits are in the air-bridge report. ## Jimmy controller conformance checkpoint User pauses gadget expansion and prioritizes Jimmy movement, preserving our playable baseline over JN-Engine's tuned movement. Integrated recovered ground/ air input and velocity, shared keyboard-turn latch, full-axis base facing before control gating, exact jetpack hold comparison and gravity intermediate.108 saved original component outputs match managed checks:25 ground,35 air,16 facing, 32 jump. REA retained complete original instructions; native component runs are separate from whole-original gameplay.27 special-contact dispatch and12 missing air-target observations are documented, with explicit implementation gates. 26 integrated movement groups,20 body groups and13 Jimmy evidence guards pass. Input-only lab and two-way underground routes pass. A pressure/capsule stair regression was caught and its evidence retained: source−250 Vy requires an actual support polygon and remains disabled in the lab/outside capsule. Native Linux 1920×1080 SDL walking and powered flight pass all five jump aliases, with resize, received-key checks, identical final readbacks per replay and zero validation errors. Native Carl/control-phase and9-stage Grappler replays also pass. Source angle normalization required signed-direction Bubble assertions; supported Grappler tests now distinguish fresh−150 versus late100 velocity publication. Continuation/coverage: [Jimmy movement](gameplay/jimmy-movement-stabilization.md). Next: ordinary-air destination Euler plus actor-root render bridge, followed by root-changing/special/dynamic contacts before broader source-body admission. The incomplete successful-climb probe is preserved in ignored artifacts, not claimed as verified. All investigator file ownership has returned to primary. No new gadget work, broad body replacement or complete movement parity is claimed. All scripts/test.sh checks now pass across the recorded run/resumptions after correcting obsolete Grappler and first-step Bubble assertions:320 Python+2 HUD, 43 managed projects and36 main Vulkan render completions with zero validation errors, asset decoders, saves/animation/audio and539 original hashes. Both packages identify cleanacd1b28 with519 matching original asset copies and4271 prepared resource files each. Standalone Linux from /tmp with SDK variables unset repeats Shift walking and right-mouse powered flight with two byte-identical development readbacks and zero validation errors. Windows native gameplay remains unverified. Logs/provenance: artifacts/jimmy-movement-audit/ and the stabilization report. ## Scooter pitch and dismount follow-up User confirmed Scooter works, then reported high pitch and Goddard retaining an angled dismount. Keep the recovered Euler handoff; add missing4096E0 full-axis leveling in the bounded Goddard follow bridge. The engine's nominal formula is correct; add original47D9D0 positional frequency correction, per-channel history and zero-distance device retention, using the shared listener eye. Source audio global scheduling/attenuation and audible Windows comparison remain open. 11 native leveling traces (9 active/2 gates) and11 native positional-frequency observations are recorded.13 Scooter/2 ScooterAudio/10 Patrol groups,24 targeted Python checks, lab/underground routes and SDL dummy pause/lifecycle pass. Both 1920×1080 native SDL Scooter replays pass10 phases with validation enabled and zero errors. Findings: [leveling](gameplay/scooter-dismount-leveling.md) and [pitch](gameplay/scooter-pitch.md). Both packages identify cleanfccc0b9; all14 Scooter resources match. Standalone Linux from /tmp passes both10-phase SDL replays;20 readbacks match development byte-for-byte, zero validation errors.539 originals remain unchanged. Native Windows/audible comparison remains pending. See scooter-live.md for provenance. ## Live Scooter in VR05 and outside LEV1 Implemented independent vehicle/wheel bodies, summon-at-Goddard and strict contact mount, normal owned-tile outside selection, drive/steer/brake/reverse/jump, impact/recovery, rider poses, default/impact camera, engine pitch and atomic cleanup/RETURN. Exact body/wheels/rider resources use source pivots and unit scale. F1 grants/selects Scooter and shows mount/speed/wheel state; the32-row Player page fits native2560×1080. Goddard has an opt-in bounded follow subset in these scenes. 53 copied-original component observations match managed formulas. Both headless and native Linux1920×1080 replays pass10 phases with real SDL input and zero Vulkan validation errors. Actual front-door head-on contact/recovery is a separate managed test: it correctly blocks throttle for four seconds. Camera source413870 passes edges=false, correcting an initial mismatch. Source/REA ledger, coverage, port boundaries and remaining work: [scooter-live.md](gameplay/scooter-live.md). Full scripts/test.sh passes303 Python+2 HUD/all managed suites and36 main GPU completions, with zero validation errors. Final11 Scooter/10 Patrol groups and 539 original hashes pass. Both packages identify clean62c07a3; all14 Scooter resource files match. Standalone Linux from /tmp with SDK variables unset passes both10-phase native SDL replays; all20 readbacks match development byte-for-byte. Post-package originals verify unchanged. Final provenance is in the live report. Original whole-game and native Windows comparison remain pending; Jimmy's lab/outside capsule and parked broad collision work are preserved. ## Live VR08 Grappler Integrated shared body/sphere contact state, full bounded base callbacks, recovered follow camera, exact rope/SWING/cursor art, selection/audio/companion lifecycle and RETURN cleanup. Process scalar memory survives scene reconstruction; noclip/ reset teardown is explicitly modern debug policy.113 copied-original component observations support the composition;13 managed live groups and40 Python checks pass. Native Linux GTX1070/private Xvfb1920×1080 input replay passes9 phases with zero enabled Vulkan validation errors; headless replay also passes.32-row Player debug page includes Grappler and fits widescreen. Details/limits and final build evidence: [grappler-live.md](gameplay/grappler-live.md). Whole-original comparison and native Windows gameplay remain pending. Both packages identify clean43806fe; all9 exact resource files match. Standalone Linux from /tmp passes the9 SDL phases with validation enabled, and all9 stage readbacks match development byte-for-byte. Full scripts/test.sh passes297 Python+2 HUD/all managed and headless route/gadget/audio checks; all539 originals verify unchanged. Final suite log: artifacts/grappler-live-final-tests.log. Package log: artifacts/grappler-live-package.log. Earlier component-only status follows chronologically. # Progress — 2026-11-03 ## Grappler ray, constraint and contact checkpoint Added boundedVR08 GrapplerRay and first-hit polygon/material capture, plus aim gating, quick-magnitude rope constraint and post-base contact response.42 new copied-original cases (23 aim/10 constraint/9 impact) match; with earlier30, 10 managed groups cover72 observations.18 Python,11 camera/13 Rocketship groups and .NET10 build pass. ActualVR08 authored-start rays hit with correct metadata; lab gameplay readback remains byte-identical and actualVR08 preview renders with enabled Vulkan validation, zero errors.539 originals unchanged. Important correction: the attach leaf's lack of duplicate material filters does not bypass aim rejection.426098..4260A0 ends active input on429D00's true return; close/rejected surfaces therefore prevent later primary dispatch. Rejected material may reel before returning; close distance does not. Strict>60 post-base contact detach and normalized source rope offsets are independently probed. Source/evidence/limits: [grappler-query.md](gameplay/grappler-query.md). No live Grappler or package update is claimed. Next: shared47B4B0 collision state and43AA40 composition, follow camera, rope/cursor render+reset/lifetime, then boundedVR08 live attach/reel/swing/detach/control-loss/RETURN. Pure constraint and post-base suffixes do not establish those integrations. Playable70cba65 retained. ## Grappler scalar checkpoint Implemented isolated GrapplerMotion attachment length/first-scale, held/fast reel, damping and jump/fire detach predicates. Thirty copied-original Wine32 scalar observations match5 managed groups;14 Python guards/projections, existing25 movement/15 gadget groups and main .NET10 build pass. Source/log/generator/binary hashes and source offsets are committed with the fixture. No live mode7, ray, attached body, rope/cursor rendering or camera is enabled. Scope and retained REA findings: [grappler-component.md](gameplay/grappler-component.md). Current playable Linux/Windows packages identify clean70cba65, including the user-clarified orbit Left/Right correction; native Linux input/resize validation passes. Grappler component source is a later local checkpoint, not a new gameplay package claim. All investigator ownership returned. Next: selected material ray context, original attached collision composition, helper rendering/lifetime and shared camera, then explicit boundedVR08/control-loss/RETURN integration. ## Orbit horizontal input follow-up and Grappler continuation User clarified the remaining orbit issue is Left/Right control inversion, with Up/Down already correct. Reversed only orbit Left/A and Right/D azimuth input. The corrected coordinate convention and vertical/gameplay input remain unchanged. Build/package validation follows in orbit-camera.md. Grappler is the next gadget. Independent investigations retained REA/source artifacts before reaching usage limits; primary owns all files again. Preserve artifacts/grappler-{ray,motion,resources}/. Important recovered leads: the attach leaf uses any ray hit, but earlier aim rejection can prevent its dispatch; VR08 slot134 returns false through475CD0; root-local ray discards the swing pivot+116. Attach uses x87 sqrt, swing normalization uses OMTquick magnitude; ordinary437940 frame still runs after the swing input short-circuit. Rope is HIROPE→rope01.ASE with jimycarl.png; cursor is retainedsprites.omt8. No live code/exporter was added. Next primary work is review/guard retained results, complete component probes, then implement boundedVR08 only when body/camera/helper contracts support it. ## Orbit camera reflection correction The modern inspector paired a right-handed view/perspective with source scenes whose gameplay cameras use positive-Z left-handed space. This mirrored the world on switching camera modes. OrbitCamera now supplies the matching left-handed view/projection and retains the shared single Vulkan Y correction.36 camera/ aspect matrix cases pass plus existing35-level projection checks. Before/after GPU captures confirm the removed reflection; lab gameplay captures remain byte-identical. Main build passes. Native/package evidence is recorded in [orbit-camera.md](orbit-camera.md). This is a port camera fix, not new original runtime evidence. Native Linux GTX1070/private Xvfb1920×1080 orbit input/resize check passes with validation enabled and zero errors. Both packages identify clean d8f1b51; final standalone Linux capture matches corrected source byte for byte.539 original hashes verify after packaging. Native Windows runtime remains pending. ## Rocketship wall-impact correction User Windows testing exposed wild ship scaling on wall hits. The initial port misread first-material RGBA writes as geometry scale. EXE43E76E selects material interface+500;43E78B..43E7A7 writes(1,pulse,pulse,1), and ourDLL10037874..100378FA consumes+4C/+50/+54/+58 as diffuse RGB/alpha. Rocketship now stores ImpactColor; visuals keep unit geometry and tint only the ship material. Driver/Goddard/fire positions, scale and color are independent. Expiry/reset restore white. 13 managed groups pass, including two actual static-wall callbacks and per-vertex geometry/color checks over130 frames with repeated renders/expiry. Copied-original numeric fixtures retain their historical raw `scale` label with corrected meaning; Python projections use color and guard the EXE writer/material constructor/DLL consumer. Main build and targeted source/render checks pass. Details, REA evidence and final package verification: [impact report](gameplay/rocketship-impact-render.md). Both packages identify clean62e908e and retain18 exact resource files. Final standalone Linux GTX1070/private Xvfb1920×1080 flight replay passes all10 SDL phases with validation enabled and zero errors;22 Python Rocketship checks and 539 original hashes pass. No current native Windows runtime claim is added. ## Live VR02 Rocketship Implemented owned/reused3ROC flight in VR02: source ceiling/wall/support profile, registration-ordered body/input/normal phases, original Euler seam/facing, Enter/T turn and R barrel roll, old-eye camera, driver/Goddard/flame art and engine loops. Mounted Jimmy retains static physics before ship-root synchronization. Pickups use controlled ship radius50; LOAD uses actual Jimmy radius60. Tab/Space, F1, cutscene cleanup, force-control, reset and noclip ownership are integrated. Source/probe details and boundaries: [rocketship-live.md](gameplay/rocketship-live.md). 96 copied-original observations cover input/vertical, motion/camera, callbacks, support and impact components.14 managed flight groups,12 final integration groups and source/projection checks pass. Native GTX1070/private Xvfb1920×1080 passes all10 render/control phases and real SDL Up/Left/Enter/R/Tab/F3, engine cleanup and zero Vulkan validation errors. Headless matches;295 accelerated ordinary ticks reuse received input without relocation or timer mutation. Captures inspected. Full scripts/test.sh passes268 Python+2 HUD/all managed/33 GPU+17 material checks; 539 original hashes verify. Initial selected-project compile dependency was fixed by passing a DirectorActor contact descriptor; failed log remains separate. Both self-contained packages identify clean e77add5; all18 Rocketship resource files/four exact RGBA hashes match. Final standalone Linux from /tmp passes all10 SDL phases with validation enabled;1920×1080 Player-page capture fits32 lines. Package logs/provenance are in the live report. Other levels, special hazards/missions/RNG, original pair ordering and native Windows comparison remain open; preserved collision/reference baselines remain untouched. ## Give All Gadgets debug action Added user-requested F1→Player→Give All Gadgets. It sets items0..7 to available ownership1 without changing selection or other inventory entries. The Player page remains within32 overlay lines. Debug-menu navigation/repeat/layout checks and main .NET10 build pass; existing NU1900 metadata warning remains unrelated. Headless llvmpipe1920×1080 Player-page render passes with zero Vulkan validation errors; readback inspected and all rows fit. Log: artifacts/debug-give-all/render.log. Normal level-load availability policy still applies; no gadget behavior changes. ## Earlier Rocketship component and Grappler recovery Historical component-only status below is superseded for VR02 by the live checkpoint above. Grappler gates remain open. Continued user-requested gadgets using supplied decompiler leads, independently checked with local REA and guarded original bytes. Integrated source/resource reports and exporters for Rocketship and Grappler; no friend gameplay code was imported. All investigators returned ownership. Details: [rocketship-component.md](gameplay/rocketship-component.md). Implemented isolated RocketFlight ordinary keyboard scalar/bank and vertical publication. Two copied-original Wine32 probes establish14 input and14 vertical references;13 managed groups compare the committed numeric fixtures and threshold/ ordering/admission cases. This is not connected live flight. Original radius50 is a collider setter, correcting the friend's visual-scale interpretation; Up/Down650 are acceleration commands, not direct velocity assignments. Exact Strato/firestrato/Jimmy/Goddard resources and VR02 authored provenance export separately with5 checks. Rocket source3/Grappler12 groups pass, along with existing Gadget15/Movement25 groups and main .NET10 build (existing NU1900 metadata warning). All539 originals verify unchanged. Native component probes are distinct from whole-game, new Vulkan/live gameplay or native Windows validation. Packages remain the prior playable d6ca417; preserved collision baselines remain untouched. Next: close aircraft body callbacks, Euler/slip/roll/camera, owned-versus-placed property/reset and passenger/effect/helper lifetime, then integrate explicit vehicle ownership and boundedVR02/F1 with live input/control-loss/RETURN tests. Grappler ray/material context, attached body/camera and helper lifetime remain separate integration gates; no nearest-hook/direct-pull substitute was enabled. ## Supplied jn-engine research assessment User requested review of the friend's SURVEY-for-Rockman.md and jn-engine-master. Survey read in full; independent source/gameplay and capture/tooling reviews are integrated in [reviews/jn-engine-survey.md](reviews/jn-engine-survey.md). Useful class/address index, saved Jimmy helper bodies (including429D00), animation and campaign leads, capture tooling and derived Level1 fixtures. All208 ledger rows are specs, not implemented/fidelity-verified classes. C gameplay uses explicit approximations and conflicting gadget modes; retain our existing C# port. EXE SHA matches the friend's declaration; OMT2.dll differs. All254 ASE/126 PNG/ 35 GAM shared files are byte-identical by case-insensitive name and SHA-256. Complete539-file original manifest verifies unchanged. Mixed Jimmy pointer bases, stale timer naming and capture-specific renderer claims require validation. No raw OMTC, analysed Ghidra database or proposed gameplay state tap is included. Prepared a local evidence-request document; no external message was sent. Documentation-only assessment: no gameplay changes, friend-tree edits/execution, network, build or new runtime claim. Both investigators returned ownership. Use the saved bodies to guide the next bounded REA gadget recovery; original state captures would strengthen future cadence/control/transition comparisons. Direct source import awaits recording the author's offered permission/licence. Existing d6ca417 playable/packages and collision gates are preserved. ## Baseball continuation Continued gadgets after clean1ae5f42/50fcb87. Local REA investigations recovered Baseball's20-slot pool, hold/release and delayed landing impulse, VR07 moving targets/strict parent activation and exact TIME font. Original addresses/offsets, fields, hashes, provider limitations and closed Evidence are in gameplay/baseball*, with separate source guards and tests. Scooter recovery is saved in scooter.md and vehicle-assets.md; its separate C3DJEEP/wheel body gates prohibit pretending that a Jimmy mode implements it. All investigator file ownership has returned. Implemented BaseballGadget/Projectile, explicit ball body response, SceneBaseballs, original ball/aim/TIME rendering, VR07 source static registration and F1 grant/ diagnostics. Pending throws/flying balls survive unequip; source scalar/label and sound/launch/cost order are separate. Six identical target tags preserve distinct instances. Review corrected gadget-specific companion restoration and hidden trophy contact/conditional activation;12 managed/6 visual and16 Baseball Python groups pass. Bounded pair discovery and angular/world/Euler/helper retention limits remain explicit; full enemy AI/combat is not claimed. Headless and native GTX1070/private Xvfb1920×1080 pass all eight rendered phases, including real SDL Ctrl/Left/Up/Tab, actual target hit, retained projectile and respawn; enabled Vulkan validation reports zero errors. Fixture advances1010 extra ordinary player/target ticks, preserves tutorial energy20 and does not relocate actors or directly modify timers. Captures were inspected. Complete scripts/test.sh passes232 Python+2 HUD, all managed suites,32 main GPU completions and17 isolated material checks with zero enabled validation errors. SDL dummy lifecycle and539 original hashes pass. A final focused12-group Baseball run adds source wall-reflection and.46 floor-friction numeric checks;16 Baseball Python groups pass after adding the font-table guard. Logs: artifacts/baseball-implementation/. No new native Windows/original whole-game runtime claim is made. Source milestone d6ca417 is committed locally. Linux-x64 and win-x64 packages identify that clean source (source_dirty=no); both publish successfully despite NU1900 unavailable vulnerability metadata. All16 referenced Baseball/aim canvases, including exact TIME glyph pixels, verify in both. Standalone Linux from /tmp passes all eight Baseball phases and a1920×1080 F1 Player view with the new grant and diagnostics; release validation layers are unavailable, separate from enabled source/native checks. All539 originals verify after packaging. Logs: package.log, package-baseball.log, package-debug-player.log and originals-after-package.log. Live ball bodies/contacts are bounded to LV1B/VR01/VR04/VR07/LEV1; other levels, full enemy combat and native Windows gadget verification remain open. Preserved baseline packages and parked collision checkpoints remain untouched. ## Bubble continuation User continued gadgets after clean1897dd2/9ea5784. Bubble mode2 motion, body, sprite and VR01 floor dependencies are recovered in gameplay/bubble-motion.md, bubble-body.md and bubble-rendering.md. Six isolated copied-original body cases establish160/.415 support66.4, no forced lowering on equip, cleanup60/1.265 and 300/900 bounce motion. REA Evidence, guarded source reports and probe fixtures are retained locally. Full original gameplay/native Windows remain separate. Primary integrated BubbleEffect/Motion/Visuals, controlled−.5dt, fire-as-throttle, persistent movementmode2, direct300/900 bounces, source cleanup650, original Canv1 full RGBA fade and F1 grant/select. VR01 now uses the recovered static body subset with retained-material111 lava/score/audio/debounce policy. A stored material owner preserves Invisible114/FC versus late lava RGBA write order. The original fixed75.9 root/render bridge remains; lab/outside still use provisional collision. GadgetScreen reads actual raw mode and Bubble practice loads before live equip. Thirteen Bubble integration groups pass, alongside11 pure motion and8 effect/ visual groups. Independent review reruns those suites and finds no blocker within the declared profile. Complete scripts/test.sh passes202 Python+2 HUD checks, all managed suites,31 main GPU completions,17 isolated material checks and SDL dummy lifecycle, with zero enabled Vulkan validation errors. All539 originals verify. Native GTX1070/private Xvfb at1920×1080 passes all eight Bubble SDL phases, including real supported900/SFX106, cleanup650 and visible full RGBA fade. Its accelerated fixture advances306 extra ordinary player ticks and retains tutorial energy20; it does not prove original cadence or manual tutorial completion. Readbacks were inspected. Logs: artifacts/bubble-implementation/. All investigator ownership has returned to primary. A20's consumer remains open; normal accepted-input target handoff is recovered without guessing a countdown. Source milestone1ae5f42 is committed locally. Both self-contained packages identify that clean source (source_dirty=no) and contain exact Canv1 pixels verified against their recorded RGBA hash. Linux/Windows publish succeeds despite NU1900 unavailable vulnerability metadata. Standalone Linux from /tmp passes all eight Bubble phases and a1920×1080 expanded F1 Player view. Release validation layers are unavailable; enabled source/native checks remain separate. Windows is cross-built, not natively gadget-tested by the agent. All539 originals verify after packaging. Logs: package.log, package-bubble.log, package-debug.log, originals-after-package.log in artifacts/bubble-implementation/. Preserved baseline packages remain untouched. No broad C4/C5 resume: dynamic/special lab/outside collision, shared camera/Euler ownership, all damage flags and retained owned-helper lifetime remain explicit boundaries. ## Invisible and shared energy continuation Implemented original item2/raw mode2: initial7-second recharge, fixed2-energy activation,7-second active timer, original alpha.7/last3-second blend flash, original Canv21/22 meter, companion hide/body flags and source cleanup. Enter/ Ctrl/left mouse activates after recharge without jump. F1→Player grants/selects Invisible, unequips it and shows remaining/recharge/targetability diagnostics. Raw-null control freezes its timers; named-Jimmy CAM/MCA DeactivateInv cleanup is independent from raw-owner preparation. Full enemy sensing remains unported. PlayerEnergy now retains the shared scalar/warning latch across ordinary loads; ray, active jetpack and Invisible adjustments use original empty80/low182 sounds with strict19/20 hysteresis and source callback order. The existing complete 42A920 REA record is reused; new independent REA reports cover Invisible, material/AI dependencies, meter art/UV/rendering and Bubble's next-step contract. All investigators have returned file ownership; original files remain read only. Fifteen managed Invisible/energy groups pass, including binding/control/companion and widescreen meter checks.182 Python checks and independent review pass. Native GTX1070/private Xvfb at1920×1080 passes real SDL Ctrl/Tab through all six rendered phases, zero enabled Vulkan validation errors; headless matches. This explicit fixture advances836 extra normal player ticks, not an original-cadence capture. Six-phase readbacks were inspected. Complete scripts/test.sh passes182 Python+2 HUD checks, all managed suites,30 main GPU completions and17 isolated Vulkan material checks with zero enabled validation errors. SDL dummy lifecycle and all 539 original hashes pass. Log: integration.log. Clean package/standalone release results will be recorded after the verified source milestone. Logs: artifacts/gadget-followup/; reports: gameplay/invisible-implementation.md, invisible.md, player-energy.md, gadget-followup-assets.md and invisible-review.md. Bubble evidence now records original Canv1 effect, equip/cleanup, controlled energy drain and movementmode2/radius/factor changes. Runtime waits for that mode's full movement/jump/body contract. Other gadgets/full enemy AI, global suppression lifecycle, graphics-option/depth state, cached canvas alignment, original runtime and native Windows verification remain open. Source milestone1897dd2 is committed locally. Linux-x64 and win-x64 packages identify that clean source (source_dirty=no). Both publish successfully despite NU1900 unavailable vulnerability metadata. Both contain exact Canv21/22 meter pixels verified against their recorded RGBA hashes. Standalone Linux launched from /tmp passes all six Invisible phases and a1920×1080 expanded F1 Player-page render. Release validation layers are unavailable; enabled source/native results above remain separate. Windows is cross-built, not natively gadget-tested by the agent. All539 original hashes verify again after packaging. Logs: package.log, package-invisible.log, package-debug-player.log, originals-after-package.log in artifacts/gadget-followup/. Preserved baseline package copies remain untouched. ## Gameplay gadgets — user-directed continuation 2026-10-27 UTC: user explicitly parked further collision recovery and chose gameplay gadgets. Preserve35a2a66/0272f7b as the collision resume point; original special/dynamic/shared-context and camera gates remain documented. The previous feature-expansion pause is superseded for this work. Independent investigators own gadget controls, resources and shrink-ray source reports/helpers/tests under exclusive names; primary owns gameplay/UI/runtime integration and root docs. The original gadget panel and combined shrink-ray/jetpack are now implemented. Tab/Space opens the original art with owned/hover tiles, Return and direct Practice selection, uniformly scaled for widescreen. Display item3 selects normal raw mode−1; raw mode0 is Bubble, correcting older inferred Scooter labels. Enter/Ctrl/left mouse fires the owned ray; jump remains the powered-jump control. Actual VR04 soldiers use original WALK/SHRINK geometry, hold the small endpoint, award10 once per hit and restore authored mode/WALK after the source timer expiry. No uniform-scale substitute, invented targets or projectile cooldown is used. Independent local REA reports retain exact source addresses/offsets, guards and Evidence for controls, resources, targeting, heading, blend and target state. Managed heading matches15 copied-original DLL references and65,540 sweep cases. The first broad suite passes157 Python+2 HUD, all managed groups,29 game GPU completions and17 isolated blend/readback checks, with zero enabled validation errors and539 unchanged originals. The complete final rerun also passes after the close-before-selection/authored-mode corrections, including15 gadget groups. Native Linux/private Xvfb passes24-frame1920×1080 menu input/capture/pause and 90-frame1280×720 ray input against actual VR04 yoke1, all three texture phases, authored55.119 mesh height, unchanged tutorial energy20 and zero validation errors. These observations verify port behavior, not full original-game or native Windows parity. Logs: artifacts/gadget-implementation/ and artifacts/gadget-blending/. Final logs: integration-final.log, native/menu-final.log and native/shrink-final.log. Scope, state owners, modern policies and remaining gates are in [gadget implementation](gameplay/gadgets-implementation.md). Full enemy AI/combat, other gadget effects, energy-warning audio and original lifecycle/visual comparison remain open. Source milestone7b5230c is committed locally. Both self-contained packages identify that clean source (source_dirty=no); Linux and Windows publish succeeds despite NU1900 unavailable vulnerability metadata. Both contain54 menu canvases and the soldier/VR soldier/ray/SHOOT manifests. Standalone Linux launched from /tmp passes the1920×1080 menu readback and90-frame VR04 ray/actual-target fixture, with authored 55.119 mesh height, all texture phases and energy20. Release validation layers are unavailable, distinct from the enabled source/native checks. Windows remains cross-built, not agent native-tested for gadgets. All539 original hashes verify again after packaging. Logs: package.log, package-menu.log, package-ray.log in artifacts/gadget-implementation/. Preserved baseline packages remain untouched. ## C4 ordinary body implementation — 2026-10-27 UTC OriginalBodyPhysics composes the source static dispatcher, positive-Y versus horizontal-first split, ordinary contact370/374, support/landing, published160 friction and suppression. All ten retained original-code body references match. VR04 now uses this source static subset; lab/outside retain the capsule pending special contacts and dynamic integration. The bridge uses actual radius60/100 for both sphere response and support, with grounded restoration in the later wrapper. Published body speed is separate from controller speed and survives script preparation/control loss until a controlled publication; Init restores authored speed (route default0). Ordinary callbacks cannot change root, so this subset permits fixedW1 publication; full callback reentrancy remains a special-policy gate. An independent investigator cross-checked source phase/flag branches, rootY clamp, radius setter/getter and authored/live field chains with REA Evidence and guarded instructions. The same investigator reviewed the implementation and independently ran twenty body groups. No directly evidenced mismatch was found within the declared zero-angular/ordinary/static subset. Other original VR04 dynamic classes, generic rotations, initial/visual root and modern noclip capsule admission remain explicit gaps. Reports: checkpoint/collision-body-composition-source.md, collision-body-implementation.md and collision-body-implementation-review.md. Full scripts/test.sh passes129 Python plus2 HUD tests, all managed groups including 20 body and11 dispatcher,27 GPU completions with zero enabled validation errors, SDL dummy lifecycle and539 unchanged original hashes. Lab→tutorial→lab→outside, RETURN/scalar/input ordering, control lock, objective/save and jetpack regressions pass. Initial suite compile failed because two explicitly linked test projects needed the new body dependencies; fixed and rerun successfully. Logs under artifacts/body-composition/ retain both attempts. After the final authored/live speed guards were added, the complete Python suite passes130 tests in python-final.log (eight body-source groups,81 sites/12 targets). The prior full integration run's129 count precedes that extra source regression; no runtime implementation changed after its successful completion. Native Linux GTX1070/private Xvfb at1920x1080 passes the300-frame VR04 SDL move/ brake/turn/jump replay and resize800x600, with zero Vulkan validation errors. Native/headless source-body runs end grounded with travel403.88, one burst, max rise418.93, energy20 and two support landings; radius-dependent support is retained instead of flattening this diagnostic count. Final readback inspected; modern sole bridge is consistent, not original pivot verification. No new native Windows or whole-original-game comparison claimed. Baseline69d0187 remains intact. Next: original dynamic actor/pair eligibility and shared reclip context, then Player160/164/FENCE/LADDER for lab/outside; C5 shared eye/DEFAULT remains separate. New package results will be recorded after the verified source milestone. Both self-contained packages now identify clean35a2a66 (source_dirty=no). Standalone Linux launches from /tmp pass the300-frame source-body tutorial flight and the nondefault lab→tutorial→lab RETURN/root/scalar/companion fixture. Release validation layers are unavailable; enabled source/native checks above remain separate. Windows is cross-published only. Packaging emits NU1900 for unavailable vulnerability metadata but both builds/publish succeed. All539 originals verify again afterward; baseline package copies remain untouched. Logs: artifacts/body-composition/package.log, package-tutorial.log and package-return.log. ## C4 static dispatcher, contacts and ordinary-body probe Independent REA investigations and primary integration recover47B4B0's reachable retry, source-order cell/FIFO duplicates, mutable nearest/last-polygon state and terminal normal callback. OriginalStaticCollision implements this bounded grid dispatcher and strict nearest support query as an isolated component. It remains outside the playable body path. Seven copied-original dispatcher fixtures match managed positions, pass/visit counts and callback polygon observations. Review corrected initial nearest-distance gating and geometry-disabled publication with retained callbacks. All investigator file ownership returned to primary. Recovered370/374/37C, landing-before-velocity-clear, published160 versus controller608 and friction/suppression. The Player material discriminator is linked3DMa archive ID, proven through OMT2.dll loader/database/chunk layout, not payload flags or a render index. Scene preparation and Mesh now preserve signed IDs/completeness; old metadata is rejected by the new source dispatcher. Actual LV1B/LEV1 contain special climb/fence/ladder IDs; VR04 contains none. Special policies remain open. Ten copied479D20/support fixtures completed in an isolated Wine32 harness with synthetic ordinary-body virtual contracts. Results cover support/settling, pending jump, falling/landing, friction versus published speed, stationary angular motion, suppression and geometry disabled. They are numeric references for the next adapter, not verification of a managed whole-body implementation. Corrected C3's earlier support-hit inference:479DF7..479E54 cancels gravity/vertical velocity when global geometry4F69E5 is disabled, not on ordinary grounded jumps. Prior harness omissions/logs and temporary approval-review usage failure are retained; after the stated reset, normal approval review succeeded and the probe completed. Full scripts/test.sh passes:122 Python checks plus2 HUD checks, all managed groups, 27 GPU completions with zero enabled Vulkan validation errors, SDL dummy lifecycle and539 unchanged originals. Eleven focused dispatcher groups pass, including actual route material coverage, legacy metadata rejection, retained context and support ties. Logs: artifacts/collision-integration/integration.log, dispatcher-tests.log, dispatch/wine.log and body-probe/wine.log. No new native Windows or whole-original game comparison is claimed. Existing baseline/release packages remain untouched. Both self-contained RID builds succeed into separate artifacts/collision-integration/ build directories. Restore emitted NU1900 for unavailable vulnerability metadata; compilation/publish succeeded. These are build artifacts, not a new package release or native Windows runtime verification. Logs: build-linux.log/build-windows.log. Reports: checkpoint/collision-grid-traversal.md, collision-contact-callbacks.md, collision-body-probe.md and collision-integration-review.md. Next: ordinary body composition against the ten references, then an explicit VR04 live subset with root/input/ceiling/RETURN checks. Lab/outside replacement additionally requires special callbacks and dynamic shared-context/reclip order. Shared camera eye, audio clock/banks and remaining plan chunks follow their existing gates. ## C4 sphere response and ceiling continuation REA investigated47BD80,479D20,47BBB0,47B4B0 and projection helpers on a generated EXE copy. Complete Evidence/exports/closed snapshots are retained in artifacts/collision-response/. Byte-guarded callgraph, nearest face/edge state, one-sided ray, lower-edge mode and separate support phases are documented in checkpoint/collision-response.md. An isolated copied-original kernel probe passes ten synthetic cases under Wine32, including wall face/crossing/departure, edge switches/lower-edge override and ceiling. Preferred-base mapping was unavailable; the fixture rebases only explicit guarded in-memory operands. No original files are patched or game entry point executed. Initial harness failures are retained. Added OriginalSphereResponse as a bounded component with original numeric-reference regressions. The live capsule remains until static-grid traversal, normal callbacks, negative-Y split, support/friction and dynamic reclipping satisfy the C4 gate. This supersedes the earlier recursive-sliding inference, not the whole solver. Integrated directly evidenced Jimmy overhead behavior for LV1B/VR04/LEV1: root+150 one-sided47C210 query, min(−0.6*Vy,−85) response, mode1 jump/fire cancellation before movement/fire. It runs without raw control. Fresh input still precedes Jimmy's wrapper, retained input follows Jimmy/fire. Camera-only debug toggling does not disable physical queries. Current capsule position response, other scenes and full sound/gadget cleanup remain explicit approximations. See ceiling-wrapper.md. Independent investigators supplied ceiling and camera byte guards plus REA batches; primary completed reports and registration guards when investigator capacity ended. All ownership is now primary. body-root-registration.md closes retained49E clearing, physical-list rebuild and repeated Init before placement; it does not invent a capsule height. camera-handoff.md proves resumed follow captures the shared scripted eye and pre-update aiming; that C5 implementation remains next. Provider-unrecognized entries/errors remain on disk beside successful Evidence and supplementary bytes. Targeted movement25 groups, original lab ceiling/query fixture, five sphere groups, camera query groups and source/projection tests pass. Both input-only lab/outside and underground passages use the same geometry queries as Program and pass with their score/task/contact assertions intact. Final scripts/test.sh passes: 104 Python checks plus separate HUD checks, all managed groups,27 GPU completions with zero enabled Vulkan validation errors, SDL dummy lifecycle and539 unchanged original hashes. Native Linux GTX1070 at1280x720 passes Carl/control,300-frame jetpack SDL input/resize and F1/F3 checkpoint tools (four commands, five scene renders), with zero validation errors. Logs: integration-final.log and native/. Native Windows and original whole-game comparison remain open. Both self-contained packages rebuilt from clean57dfcc6 (source_dirty=no). Standalone Linux launches from /tmp pass the control/body/fire fixture and nondefault tutorial/RETURN scalar/root/companion/availability assertions. Release validation layers are unavailable; enabled source/native-window checks remain separate. Logs: package.log, package-control.log and package-return.log under artifacts/collision-response/. Preserved baseline packages remain untouched. Windows is cross-published only. Next: complete C4 traversal/normal callbacks/ support/friction/dynamic reclipping, then shared viewport eye/aim ownership. ## Uncontrolled-player implementation and REA Continued C3 into source control/body/fire boundaries. Original475CA0/frame/body eligibility is separate from controlled pointer5099E4. Movement437940 has an identity gate after base/timer work;421FB0 fire has no raw-control gate. CAM, initial MCA and AI preparation order differ; later MCA shots do not repeat it. Source-byte reports/helpers/tests: uncontrolled-player.md, uncontrolled-physics.md and cutscene-preparation.md under checkpoint/. Investigators supplied independent player/control traces; primary completed physics and preparation work when agent capacity ended. All file ownership is now primary. Implemented LabSimulation controlled-input/controller gates while existing body physics and owned fire continue. Preparation clears speed/velocity only; queued impulse/acceleration and mode/jump/hold/fire are independent. Ordinary grounded radius60 is outside the movement gate. PlayerSceneBinding connects ordered raw preparations and authored AI root/yaw; STOP observation follows physical clear. An initial MCA with first target none returns instead of executing completion. Force Control stays an effective override; preparations use authored ownership. Headless story pre-roll uses the same phases; F1/practice/inspect still pause them. Full wrapper/base transforms/ceiling response, other gadget/class effects and registry/cadence remain explicit gaps in control-phase-implementation.md. Targeted movement24 groups, coupled director tests and the32-frame airborne control fixture pass. Native Linux GTX1070 at1280x720 passes Carl/control phases, F1 menu/Force Control and F3/tools/checkpoint reload, with zero enabled Vulkan validation errors. The input-only lab replay now uses the actual binding; its old coasting turn missed the narrow exit. Down braking before large turns reaches outside with score150, one practice No, owned jetpack, zero jumps and6 wall hits. No physical response/task gate was changed for that replay. A corner snapshot is retained for C4. Logs: artifacts/uncontrolled-phases/. REA MCP connected but lacked native providers. Installed the verified official Ghidra12.1.4 ZIP locally and extracted matching full Ubuntu JDK21 packages under .tools (system Java21 was runtime-only). Existing Node24/public REA6.8.0 launcher is selected by ignored .tools/rea.env and scripts/rea.sh. Client registration was not changed. Local Unix IPC needed the approved execution allowance. Five source procedures were decompiled and byte-cross-checked; complete Evidence/snapshots are on disk.421FB0 maps correctly but has no recognized Ghidra procedure in this profile; existing byte/vtable evidence remains authoritative. See docs/rea.md and the control-phase report's Evidence IDs. No original game execution or new native Windows runtime comparison is claimed by this setup. Final C3 validation: full scripts/test.sh passes (integration-final.log):93 Python checks, all managed groups including24 movement, updated input-only route, 27 GPU render completions with zero enabled Vulkan validation errors, SDL dummy lifecycle and539 unchanged originals. Coupled director tests include body-before- STOP observation and the initial empty-MCA return. Four native window commands pass (five scene renders) in native/*.log. No native Windows runtime is added. Both self-contained packages rebuilt from clean commit c285b57, with source_dirty=no. Standalone Linux launches from /tmp pass the airborne control/body/fire fixture and nondefault lab → VR04 → lab retained-state/RETURN checks. Exact root, companion restoration and jetpack availability also pass. Those release runs use llvmpipe with validation unavailable; the enabled source and native-window validation above remains separate. Logs: package.log, package-control.log and package-return.log under artifacts/uncontrolled-phases/. Preserved baseline packages remain untouched. Windows is cross-published only. ## Retained-state implementation continuation Read prompt.MD in full and resumed C3/C7. Three independent investigators owned new property-reader, RETURN-cleanup and companion files; all ownership returned. Reports: checkpoint/retained-streamer.md, retained-return-cleanup.md, retained-companion.md and retained-player-body.md. Exact source offsets/targets, structures, callers, scope and unresolved fields are recorded with14 new Python regressions. Generic Init re-applies authored body vectors; final preparation clears speed/velocity XYZ; jump state resets independently of retained scalars. Implemented bounded PlayerLevelState (mode, jump-hold, owned-fire timer) for ordinary LV1B/VR04/LEV1 transitions in the supported no-gadget/jetpack cases. Other body/debug state reconstructs separately; TSK/debug checkpoint loads do not carry transient fields. Modern R reset clears them. Companion AI/target/waypoint/ rotation and shadow resource resets are documented, not blanket-persisted. Corrected prior prefix-only gadget availability: full VR01..08 PostLoad tails re-enable the matching practice gadget; owned VR04 jetpack finishes1, not4. LV1B disables gadgets1/7. Unknown wire states and other inventory regions stay intact. Other level-specific gadget tails remain explicitly incomplete. Older VR04 state4 claims below describe the superseded partial implementation. Targeted22 movement groups and inventory tests pass. Isolated original routines pass under Wine32 in a copied, filesystem/network-isolated installation: authored vector leaf accessors and43A750 preparation preserve unrelated sentinel bytes. Synthetic fixture vtables resolve to guarded source targets; no whole-loader, original-game end-to-end or native Windows verification is implied. Fresh Wine prefix setup failed; a copied existing generated prefix works. Probe logs and build/integration artifacts: artifacts/retained-lifecycle/. Validation:84 Python checks (14 new), all managed groups including22 movement, 26 GPU render checks with zero enabled Vulkan validation errors, and the final nondefault lab→VR04→lab boundary replay pass. Final boundary logs show mode1, hold.625/fire6.125 retained independently of jump0/velocity0/speed0 at both loads; RETURN position/control/companion and jetpack availability1 also pass. The full suite's original-effects smoke initially failed its fixed200ms sampling window. It now creates both streams paused before sampling and polls consumption with a 3s bound; all four final SDL dummy audio checks pass separately. No gameplay audio lifecycle was changed by that harness correction. Logs: integration-final.log, route-final.log, audio-final.log, evidence-final.log and hashes-final.log. All539 original files verified unchanged. Build succeeds with zero warnings/errors. Both self-contained platform packages rebuilt from clean9ad5a8e (source_dirty=no). Linux release launch from /tmp passes both nondefault retained-scalar boundaries and exact RETURN/control/companion checks. Release validation unavailable, unlike the enabled source GPU checks. Package evidence: package.log and package-route.log in artifacts/retained-lifecycle/. Baseline packages remain untouched. Native Windows runtime/original whole-game comparison are not added by cross-publication. ## Requested debug expansion F1 now has General/Player/Systems pages (Tab), preserving existing General actions. Added F3 noclip with vertical/strafe/boost controls, entry restoration and static exit-overlap fallback; queued motion and held inputs clear at toggles. Flight bypasses collision/camera obstruction and contacts without changing script control ownership. Added movement/input-order/availability, objective/practice, camera query, category audio mutes and original-format checkpoint actions/diagnostics. Mutes use SDL stream gain so playback clocks continue. Scene/reset lifetimes and checkpoint limits are documented in debug-menu.md. This user-requested debug work is a scoped exception; evidence-led gameplay continuation remains the direction. Full scripts/test.sh passes:70 Python,21 movement groups, remaining managed/GPU/ SDL checks and539 unchanged originals. Native Linux GTX1070 debug-tools smoke passes at1920x1080 with zero enabled Vulkan validation errors, including flight during a source cutscene without force control and same-window checkpoint reload. Widescreen/ultrawide captures fit. Logs: artifacts/debug-expansion/. Windows SDL3 exports the gain API; native Windows runtime remains unverified for these changes. Both self-contained packages rebuilt from clean d783898 (BUILD-INFO source_dirty=no). Linux standalone launch from /tmp passes Player page/noclip at1920x1080 on llvmpipe; validation is unavailable in that release invocation. Native enabled-validation evidence above remains separate. Package logs: package.log and package-smoke.log under artifacts/debug-expansion/. Baseline checkpoint package copies untouched. ## Evidence-led implementation continuation Old level-game removal is resolved by `checkpoint/level-game-lifetime.md`: class defaults briefly clear live+139, but normal Init restores it from stored+11=1; the destructor explicitly removes the registry node. Jimmy/Goddard instead store 0 and remain retained/reusable. Three new immutable-source regressions guard it. Implemented controlled player prior-body physics → fresh input/player/fire, or retained player/fire/input after scene continuation. Jump impulses wait for the next body pass; input uses stored mode, and grounded release clears turn inside the player callback. Current fixed60Hz, capsule/grounding, uncontrolled-body and other grouped scene phases remain provisional. Twenty targeted movement groups and lab/underground routes pass; GPU lab→VR04→lab verifies retained input order. Implemented `OriginalCameraRay` for LV1B/VR04/LEV1: source static grids, FIFO wall-first hit, individual segment kernel, source eye weighting/gain and one query per input callback. Eleven camera groups and existing13 support groups pass. Camera shared-state handoff and full follow/input modes remain open. Implemented PostLoad gadget availability1↔4 for the recovered eight group0 items; tutorial entry/normal return, unknown values, untouched regions and TSK round-trip are tested. VR04 bypass remains. Arbitrary retained mode/fire state is not carried without the remaining property/control callback proof. Details and exact limits: `checkpoint/player-phase-implementation.md`, `retained-character-reset.md`. Full `scripts/test.sh` passes (artifacts/implementation/integration-final.log): 70 Python checks, all managed groups including20 movement/11 camera/13 support, GPU route/story/practice/RETURN/jetpack/animation with zero validation errors, SDL dummy lifecycle and539 unchanged original hashes. RETURN smoke confirms retained input ordering and enabled gadget state1 after tutorial state4. The preserved baseline's240-step walk and new phase implementation have essentially identical completed travel/final position, with the intended within-frame delay verified separately. No native Windows/original runtime comparison is claimed. Self-contained Linux/Windows packages rebuilt from clean09cf117 (`artifacts/implementation/package.log`, both BUILD-INFO source_dirty=no). Linux release runs from `/tmp` pass lab→VR04→lab RETURN assertions and a direct outside scene load; source static query construction succeeds in all three route profiles. Release invocations lack validation layers; the full source integration above runs with validation enabled. Later changes only record this package evidence. ## Architecture reverse-engineering checkpoint User direction supersedes feature-expansion priorities below. Full `prompt.MD` read; current code/findings/progress audited. Clean playable69d0187 is preserved by local tag `baseline-re-checkpoint-2026-10-25` and copied Linux/Windows packages under `artifacts/re-checkpoint/baseline/`. All539 originals verified unchanged. The existing Gitea remote configuration was left untouched and unused; older claims below that no remotes are configured are stale. No remote work occurred. Canonical continuation: `checkpoint/README.md`, `coverage.md`, `implementation-plan.md` and the three system reports. New evidence separates viewport physics/object update/sync from engine logic/morph timing, identifies construction-order registry traversal, objective-before-task assignment and per-actor Carl callbacks, DEFAULT camera snap, bank/voice ownership gaps and TSK queued/duplicate-callback lifecycle. No original full-game runtime verification. Initial architecture checkpoint exit is recorded in `checkpoint/review.md`. C1 objective-before-records and C2 per-actor Carl callbacks are implemented; the full integration suite passes, with zero GPU validation errors and539 hashes unchanged. C3 fresh registry/input bridge, C5 first-hit static ray, C6 software duration/repeat clock and C7 selective ordinary loader follow-ups are documented. All investigator ownership returned. Exact old level-game registry eligibility and retained-field reset traces gate scheduler integration; camera/audio/physics chunks have independent concrete evidence gates. See the execution ledger. Linux/Windows self-contained packages rebuilt from clean531883a; Linux release Carl held-input smoke passes from `/tmp` (validation unavailable there). Later loader defaults-bridge provenance corrections pass five targeted byte checks and do not change runtime. Package and test logs are in artifacts/re-checkpoint/. **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. ## Persistent window/renderer and widescreen UI Main now owns one RenderSession. Per-level Run replaces scene resources while retaining the SDL window and Vulkan core; resize/fullscreen replace size-dependent targets. Window bounds, focus and fullscreen survive loading. Invalid replacement validation retains rendering. Per-level audio/gameplay still rebuild; fades remain unfinished. UiViewport scales art/glyphs uniformly with bottom/right HUD anchors, centered prompts/F1 and top-left F2. Pointer mapping handles logical/drawable size, fractional DPI rounding and strict source boundaries. Nine aspect/resolution fixtures and high-DPI checks pass. Full suite passes with 539 unchanged originals; final targeted viewport checks pass. Native GTX1070: six loads, one window/renderer, stable native/core IDs and bounds/focus/fullscreen, restored windowed bounds,4:3 through32:9/fullscreen3440x1440, native Yes/No mapping, rejected-load retention and zero validation errors (artifacts/window-transition-final.log). Captures were visually inspected. F1/force and F2 input checks still pass. Actual headless lab/VR04/lab retains one renderer/revisions1/2/3 and exact RETURN root/ companion at1280x720. Both self-contained packages are rebuilt. The packaged Linux 1920x1080 lab/VR04/lab check also retains one renderer and exact RETURN state (validation unavailable in that standalone invocation). No new Windows runtime check is claimed. Details: presentation-and-ui.md. ## User-requested live debug overlay F2 toggles a compact, non-pausing live diagnostics panel. F1 also offers Live Overlay and takes display priority while open. The setting/timing object belongs to the session, preserving visibility across scene rebuilds and resets without writing original saves. --debug-overlay enables startup visibility. The12 rows expose FPS/frame time, root/velocity, steering/pose/ground/energy, control/force/pause, tasks/score/jetpack, actual camera/far plane, active cutscene and counters/errors. Source camera and orbit projection values reflect rendering. The existing font/overlay buffers are reused with a smaller content-sized panel. Targeted model/renderer/diagnostics groups pass. Native Linux GTX1070 SDL checks pass live movement77.58 travel with visibility/menu/resize checks all true and zero Vulkan validation errors (artifacts/live-debug-window.log). The prior F1 force-control check still passes. Headless GPU output was visually inspected. Complete scripts/test.sh passes, including41 Python checks, all managed/gameplay/ GPU/audio regressions and539 unchanged originals (artifacts/live-debug-full-tests.log). The lab -> VR04 -> lab headless check retains enabled diagnostics with frame samples1/2/3 and zero Vulkan errors (artifacts/live-debug-return.log); exact RETURN root/companion placement still passes. Final authored-camera inspection also renders with zero errors, and targeted camera-mode/layout/timing checks pass. Both Linux and Windows self-contained packages are rebuilt successfully. No new Windows/Wine runtime check was run. ## User-requested F1 debug menu Follow-up: Force Control is now a persistent TRUE/FALSE menu toggle. It overrides effective ownership while preserving authored control state; switching it off restores that state. Reset/new scene clears it. Cutscene execution continues and DEFAULT music restoration uses authored ownership, avoiding a debug-induced silent return. Menu/director/interaction checks pass; the extended16-frame Linux SDL test records actual forced movement during active neutron1a,14.00 total travel, all menu checks true and zero Vulkan validation errors (artifacts/force-control-window.log).539 original hashes remain unchanged. Both executables are republished for this follow-up; no Windows runtime check is claimed. The full-suite evidence below is from the preceding menu build. F1 opens a modern debug overlay with live player/scene/task diagnostics. Keyboard actions cover camera, cutscene skip, energy refill, jet-pack grant, staged SCENE editing,33 prepared levels/original starts, reset and resume. Gameplay, cutscene clocks and SDL audio pause while open; input is captured through release after closing. No new asset/dependency or original-game behavior is claimed. Targeted menu/overlay and actual SDL pause tests pass. Native Linux GTX1070 16-frame SDL test passes pause, actions, Escape/held-input capture, resume and 800x600 resize with zero Vulkan validation errors (artifacts/debug-window.log). Headless menu GPU readback was visually inspected. Complete scripts/test.sh passes, including41 Python checks, all managed/gameplay/GPU/audio regressions and539 unchanged original hashes (artifacts/debug-menu-full-tests.log). Both platform executables are republished successfully. No new Windows/Wine runtime check was run, consistent with the user's earlier preference. See debug-menu.md. ## RETURN placement and underground passage Recovered and integrated the next lab checkpoint: - TSK StoredVector is the departure root at Jimmy+9A8. Ordinary transitions capture the live root; RETURN preserves it, restores Jimmy and Goddard to it, retains destination STRT yaw and suppresses the initial STRT trigger once. Fresh TSK loading does not imply pending RETURN. Pickup42AF00's separate +1E38/+1E48 checkpoint remains unconsumed rather than being conflated with it. - Original shared level loaders omit NC-prefixed SHAPE names from terrain registration. Both player terrain and source actor support now use that rule. All2869 lab render/source triangles remain;2851 are collision-eligible, including all325 null-material proxy triangles. No material-name water filter was added. - The underground replay starts at the original riverlab STRT with explicit SCENE65 fixture, then uses only live movement/contacts. It operates both switches, crosses fully clear bars twice, waits for reclosure and checks the closed box blocks a source-gated negative sweep.2763 steps,4153.21 travel, no jumps/warps. The former85.89-unit first-step rise onto decorative water is now rejected. The headless lab -> VR04 -> lab RETURN check restores the actual departure root (4921.8735,76.49398,502.37677), separate from STARTEXP, with no intro playback. Complete scripts/test.sh passes:41 Python checks,18 movement groups,11 collision groups, four source/checkpoint groups, director/route/underground regressions, headless Vulkan scene/RETURN checks, audio lifecycle and539 unchanged hashes. Evidence: artifacts/return-underground-tests.log. The three-scene RETURN check includes exact companion placement and reports zero Vulkan validation errors. Both self-contained executables are republished successfully; no new Wine/native Windows runtime check is claimed, following the user's preference to skip that check. Reports: analysis/runtime-checkpoint.md, water-collision.md, underground-route.md. The underground fixture does not establish progression into SCENE65; full water hazards, Goddard follow/support phase, fades and original runtime comparison remain incomplete. Native Windows testing of these new changes is not agent-verified. ## User-reported lab regressions (preceding milestone ca08c46) 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 read mode1's log branch as an early return; it actually falls through to439900. The port now uses the original airborne Up acceleration, Down braking and Left/Right turn-rate rules for jumps and jet pack bursts. Source byte checks, 16 movement groups,14 vertical burst groups and the full input-only lab route pass. VR04 GPU replay travels403.88 horizontally during powered flight, turns and lands with zero validation errors. The rebuilt Linux package and Windows package under isolated Wine64 pass300 SDL-input frames with Up/Left/Down/Shift and resize800x600, matching403.88 travel. Linux GTX1070 validation reports zero errors; Wine validation is disabled. Native Windows retesting remains unverified by the agent. Details and remaining fidelity limits: analysis/air-steering.md. ## Latest user-tested correction User testing found background music continuing during cutscenes. Recovered CAM 416776..4167E1 stops/unloads the selected music; DEFAULT447F12..447F45 restarts it when control returns. The runtime now performs that handoff. Source-byte/director checks and actual SDL dummy streams verify silence through all16 lab cues, normal completion restart, reset silence and skip restart. See analysis/sound-effects.md. Both platform executables were republished with this correction. The packaged Linux and isolated Wine64 Windows executables pass the full music-lifecycle audio check. Linux packaged prompt/input/resize still passes with zero Vulkan validation errors. Native Windows listening remains unverified by the agent; source user report and exact static rules support this fix. An additional managed replay now walks through explanation, jet pack, Carl and outside LOAD using only movement inputs, source live contacts and NPC patrol/body updates. The current replay reachesSCENE60/score150 with jet pack owned after887 movement steps and7078.63 travel; the player is never repositioned. This strengthens port lab navigation coverage without proving original collision/cadence fidelity. ## Current integrated behavior - Default play runs LV1B STARTEXP and its16-cue CAM/MCA/AI camera/speech sequence. Escape/Space skips actions and restores control; `--inspect` retains inspection. - Original walking/backpedal/turn/jump arithmetic, source clips, chase camera and fixed60Hz port simulation. Powered jumping now translates the recovered state table,8-second effect timer,60/100 contact radii and energy/VR immunity rules. - Original Carl/Goddard morphs, sprite quads and9 lab model actors, including the OMT phones, source buttons and moving bars. Goddard's unregistered STOP/WALK/TALK requests retain the current source clip rather than invented aliases. - Shared patrol states1/3/7, state9 base motion and Carl's WALK/STOP profile. Body support uses recovered radii, factors, quick magnitude and one-sided trace; full wall response and Carl sensing remain incomplete. - Lab explanation -> SCENE35 -> jet pack pickup -> SCENE50 -> Carl contact/talk -> SCENE60 -> outside LOAD dispatch. Pickup adds50, Carl callback adds100 before dialogue. Source task gates/objective selector and inventory slot0/3 persist. - 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 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). RETURN restores the saved departure root; the pickup checkpoint is separate. - Button contact/cooldown/down/up/next-trigger and vertical-door duration, reversal/hold/next-trigger logic. Source dynamic-box broadphase/triangle ordering is translated; narrowphase/response still uses a modern approximation. - Ordinary stars/trophy, score and acquired flags; all12 TSK files round-trip. CLI checkpoint load/write preserves unknown fields, inventory, tasks, energy, objective selector and RETURN context. Complete runtime/menu saving is unfinished. - Music and speech use independent SDL streams. Positional effects/banks and physical audio verification remain unfinished. Original default-bank pickup/button/ moving-door one-shots and owner loop stop are now connected; attenuation/volume policy and jetpack sound callbacks remain unfinished. ## Source availability and evidence All539 original hashes remain the baseline. All35 GAM files are supplied and33 have matching OMT geometry. Level3B/E GAMs exist, each1777 bytes, and are identical after their four-byte level identifiers. Their classes name absent OMT archives, but no serialized incoming GAM transition or supplied save selects them; loader null handling does not establish a fatal failure. Normal gameplay use is unverified, so these are not an established blocker and no extra files are currently requested. The catalog preserves absent geometry without substitution. Detailed recovery: analysis/lab-progression.md, scene-entities.md, scene-models.md, patrol.md, original-collision.md, dynamic-collision.md, buttons-doors.md, ui.md, jetpack.md, level-resource-mapping.md, and save-format.md. Addresses, source offsets, confidence and approximation boundaries are retained there. ## Verification Managed checks cover source data, director/actions/cameras, task/interactions, patrol, recovered support, moving-box collision, jetpack states and HUD composition. 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. Headless llvmpipe runs render the story, model actors, practice prompt, lab exit and VR04 entry with zero Vulkan validation errors. The300-step VR04 powered replay jumps once, rises418.98 units and lands; VR immunity retains20 energy. The corrected ordinary 240-step replay travels1264.55 and rises137.62; port landing-count telemetry changed with the explicit state table. These measurements verify port behavior only. See artifacts/full-integration-test.log and practice/jetpack logs for current checks. The full scripts/test.sh integration passed at c929f94:35 Python regressions,2 targeted UI checks, all managed groups, GPU story/route/practice/VR04 replays, SDL dummy streams and539 unchanged original hashes. The subsequent music correction passes targeted source-byte/director/SDL audio checks on both packages, plus native Linux window and input-only managed navigation regressions; hashes were reverified. Both self-contained packages rebuilt successfully. Linux package runs the12s story without SDK/source/Original mounted, zero Vulkan validation errors. Its GTX1070/Xvfb practice window pauses, receives N, resumes11 steps and resizes800x600 with zero validation errors and original pickup audio. The Windows package passes the same12-frame practice/input/resize/stream check under isolated Wine64, matching simulation statistics (validation disabled). It also renders the12s story under Wine. Generated package logs live under artifacts/package-integration.log, linux-package-integration/, package-practice-window.log and windows-practice-package/. Initial native Windows loading/rendering remains user-confirmed for the earlier checkpoint; current native Windows story/input/audio is still unverified. Wine is preliminary compatibility evidence only. ## Next work and limits No usable original gameplay reference has been captured: isolated Wine GL/Vulkan attempts fail rendering. Continue static recovery and isolated DLL probes locally; comparison on a working original installation remains required for fidelity. Next priorities are progression from outside into later lab task states; original collision47BD80/479D20 and phase/cadence; audio-bank/SFX mapping; practice highlights/energy gauges; actor sensing/animation completion. Then expand remaining gadgets, damage/combat/enemies, menus/configuration, full saves and all level progression. Lighting/fog/blend/culling and sprite/pivot fidelity are open. Do not claim complete gameplay based on direct-contact route or GPU smoke tests. All previous agent file ownership has returned to primary; investigators that reached usage limits are not dependencies. Generated output stays separate from Original. Existing remote configuration remains untouched and unused. Read tasks.md and Git status before continuing; preserve source evidence and update these notes.