Scrollable captures from the running WebGL viewer
Most entries are unedited captures from host Chrome; source-data diagrams are labeled. This WebGL reconstruction loads the L4RA scene and supports driving, but the visual match and complete Bayview assembly remain unfinished. A current Route 450 scan found no placed-object/road intersections within 100 units of that start; that local result is not a citywide placement check. The combined eight-stream scene remains experimental: exact shared meshes are deduplicated, but a unified city route/traffic graph and broader placement review are still missing. Read the latest placement audit for evidence and limits. These images show the reconstruction or extracted data, not footage from the original game.
A six-second W hold moved the 240SX about 47 m through the open aisle between pallet stacks and warehouse structure. No road departure or building-collision flag appeared. This confirms a drivable route through this local area; it does not prove the warehouse placement or close camera view matches the game. The source lane texture here has no center stripe.
Open full-size capture · Read the drive measurement and limits · Open this route
A second lane-filtered scan flagged XB_WAREHOUSECLOSEA against an RDP_LANEC pavement triangle. The route view shows an open industrial corridor with the warehouse along one edge. The overlap warning alone does not prove a misplaced building, and this view does not establish the retail use or complete placement of the surrounding port assets. No source transform was changed; Chrome reported no page or console errors.
The triangle scan flagged an edge of XB_CUSTOMGEMINI_1A_JM_00 against a marked RDP_LANEB surface. From above, the curved boulevard's marked driving lanes remain open beside the storefront and overhang. This one view does not certify the entire placement or the rest of L4RA; no source transform was changed. Chrome reported no page or console errors.
The normal yaw-rate cap is now 1.25 rad/s and the turn-acceleration cap is 26 m/s², up from 1.15 and 22. This makes the low-speed yaw cap about 9% higher and the speed-limited radius about 18% tighter. In a short browser check, a 0.35-second full-right input while accelerating changed heading by 21 degrees at 3.4 m/s. The car stayed on the marked road; Chrome reported no page or console errors. These are viewer playability estimates, not retail physics values.
Open full-size capture · Read the measurement and limits · Try this drive
The 240SX paint is visibly plain even though all four body batches use the source factory silver, `METPAINTSILVER`, and `GLOSS=128`; the panorama, local environment cube, and six local lights are loaded. Chrome reported no page or console errors. The image isolates the remaining issue to the paint equation and unresolved retail draw inputs; no guessed gloss change was made.
Open full-size capture · Read the draw-state evidence · Review the earlier shader A/B
All 393 embedded images across the 29 selectable stock cars, 43 ambient traffic bodies, and the additional 240SX GLB passed decoded RGBA equality checks. These 73 files now use lossless WebP and are 3.2% smaller overall. Chrome loaded the packaged 240SX and requested traffic models with HTTP 200, reached “Traffic preview ready,” and reported no page or console errors. This improves delivery while leaving the known retail appearance and behavior gaps unresolved.
Open full-size capture · Read the size and runtime details · Try the packaged city
After 2 seconds of throttle, the 240SX reached 10.4 mph. A 0.9-second full-right input changed heading by about 63 degrees and ended at 8.8 mph, with the car still on the authored road. The current 1.15 rad/s yaw and 22 m/s² turn-acceleration limits are viewer playability estimates, not recovered retail physics. Chrome reported no page or console errors.
Open full-size capture · Read the measurement and limits · Try this drive
The package converts all 5,867 embedded city images to lossless WebP only after decoding the result and verifying every RGBA pixel against the PNG. Across the eight regional scenes, chunk data fell from 470.9 MB to 402.9 MB (14.4%). The packaged L4RA scene reached its rendered drive view with 12 ambient vehicle models and no browser page or console errors. This confirms texture decoding and startup in host Chrome; it does not establish full retail visual or gameplay fidelity.
Open full-size capture · Read the packaging and runtime details · Try the packaged city
The viewer reports the source factory silver #747270, METPAINTSILVER, and GLOSS=128. Its LOC4DYNTEX_2414a01e panorama, local dynamic environment cube, and source shadow are loaded; the ready view has 12 traffic models and no browser errors. This shows the flat appearance is not caused by a missing environment upload, but it does not prove the cube contents or lighting equation match retail. The normal renderer remains in use.
carHarmonicsNormalization=image-defaultsBoth 800×600 captures use the same route, car, camera, and decompiled paint shader. The image-defaults branch lifts the rear body only slightly; it still lacks convincing city reflections and remains much darker than the normal drive renderer. Both runs reached the rendered view with 12 traffic body models and no page or console errors. This branch stays opt-in because it does not recover the retail draw-time normalization or material flags.
Open fallback-normalization image · Open image-defaults image · Read the comparison detailsThe current 800×600 WebGL2 capture reached the rendered drive view with 12 ambient body models available and no page or console errors. The opt-in retail-fallback path renders, but the silver body is darker and still lacks convincing city reflections. I left the normal viewer shader as the default; the exact live paint selection, texture binding, and harmonic transform remain uncaptured. This is a current-build check, not a same-frame comparison; the earlier controlled A/B pair is r415.
Open full-size capture · Read the current shader check · Open the earlier default capture · Open the earlier fallback capture
The draw wrapper passes 3.0 to the retail bounds helper, not 4.0. After updating the equation, all six visible preview cars still report a proxy metric of 0 and select the last detail level; the upright cars remain visible and Chrome reports no page or console errors. The retail wrapper culls metrics below 4, so this confirms the viewer's BODY_A bounds sphere and normalized scale are not enough to reproduce retail visibility. I have not added that cull because it would remove these currently visible cars. The live render radius and axis weights still need to be recovered.
Open full-size capture · Read the executable and geometry evidence
After raising the ordinary yaw cap to 0.95 rad/s, turn acceleration to 16, and visible wheel lock to 0.72 rad, a controlled one second throttle-and-left input changed heading by 0.70 radians over 3.0 metres, ending at about 10 mph. The car remained on the road in the inspected capture. Chrome reported no page or console errors. This is a viewer handling estimate, not a recovered retail turning curve.
Open full-size capture · Try the same city start
The full L4RA triangle scan flags many road crossings, but its predicate also catches building edges and foundations at pavement boundaries. This focused route 716 view shows a narrow cobbled street with buildings close on both sides; it does not show an entire building across the drivable lane and does not prove citywide placement correctness. No source placements were changed. See the full-stream intersection counts and method limits.
Open full-size capture · Open this route view
The draw wrapper passes a 3-unit near value to the helper: it returns twice the radius inside 5.25 units and otherwise floors radius × 3 / (distance − 3). In the usual wrapper branch it then applies a 0.7 scale before the 4-unit cull and the 40/20/10 detail thresholds. This capture predates correction of the near value and equation in the viewer. All 36 available A/B/C traffic bodies loaded, six visible cars selected BODY_C, and all six were upright and road-supported. The retail radius and weighted transform axes remain unknown, so this capture does not verify exact distance selection.
A controlled host Chrome drive exposed an overly abrupt previous tune: a 1.8 second throttle-and-left input changed heading by about 2.57 radians over 4.2 metres. I reduced the ordinary yaw cap to 0.7 rad/s and turn acceleration to 12. Repeating the same input changed heading by 0.94 radians over 5.6 metres; Chrome reported no page or console errors. This remains an approximate viewer handling model, not a retail physics match.
Open full-size capture · Try the same city start
Host Chrome uploaded all 223 nearby textures and reached “Traffic preview ready” in about 18 seconds, with no console errors or failed requests. This startup view shows the player car on the marked roadway and upright traffic. It is a single stationary view; it does not establish a match to retail driving or complete city fidelity. The ordinary steering tune is now tighter; the BODY_A/B/C mapping is verified, while exact retail LOD radius scaling remains open.
Open full-size capture · Open this city drive
The packaged default had regressed to pavement gain 1.0 despite the earlier same-camera comparison documenting 1.6. The restored value brightens the mid-road crop from mean RGB 98/101/102 to 122/126/127; player and building materials are unaffected. The yellow double line and white lane dashes remain sampled from the original road textures. Chrome reported no page errors.
Open baseline · Open restored view · Read the comparison · Try the city drive
The viewer-side yaw cap increased from 1.2 to 1.4 rad/s and turn-acceleration cap from 11.5 to 13.5 m/s², reducing turn radius by about 15% where the acceleration cap applies. A 0.9-second full-right input changed heading by 73 degrees from 13 mph; the final capture shows the 240SX on the marked road at 4 mph. The road-support speed guard also affected the result, so this is a playability check rather than a retail handling match. No page or console errors were reported.
Open the full-size capture · Read the measurement and limits · Try this drive
The road, 240SX, controls and route data now clear the loading cover without waiting for all 12 optional traffic bodies. With only traffic GLB requests slowed by five seconds each, the viewer reached “City scene ready” while traffic remained 0/12. After a two-second W hold, the car was moving at 13 mph with original AEMS/Ginsu audio active and the stream indicator at 1/12. City-texture and traffic progress no longer hide one another. Chrome reported no page errors or failed requests; traffic paint and retail AI remain unfinished.
Open the full-size capture · Read the startup trace · Open this drive
The normal yaw cap increased from 1.1 to 1.2 rad/s and turn-acceleration cap from 10.5 to 11.5 m/s². In a host Chrome capture, the 240SX changed heading by 26 degrees during a short input and remained supported by the authored L4RA road surface; no road or building collision state was reported. These are viewer playability values, not retail physics. A separate live HUD readback confirmed the speed number tracks player speed (3.03 world units/s displayed as 7 mph) and the RPM readout updates. Body paint and the wider city placement review remain unresolved.
Open the full-size capture · Read the control check · Try the same drive start
Host Chrome/WebGL2 captured the same Route 736 parcel van and camera with the normal fallback and the opt-in decompiled shader branch. The DEFAULT diffuse, specular and environment values only change shading slightly; the body remains an almost featureless white box. The missing explicit traffic descriptor state, material flags, paint mask and draw-time bindings are still required. The normal renderer is unchanged; this diagnostic is not a visual fix.
Open baseline · Open diagnostic · Read the result · Open the opt-in view
The live selector exposed and loaded all 29 stock cars in one L4RA driving session. Each selection reached its ready state and showed that car’s matched original AEMS/Ginsu recordings; Chrome reported no page errors or failed requests. The pictured 350Z moved from rest to about 7.0 m/s (16 mph), reached 2,560 rpm, and reported audio active. This verifies runtime selection for the roster and driving for the pictured car, not a separate driving or appearance evaluation of every model. Retail handling and the flat traffic appearance remain unresolved.
Open or save the capture · Read the 29-car live selector audit · Drive the packaged city
The current packaged build raises the viewer-side normal turn acceleration to 10.5 m/s² and yaw cap to 1.1 rad/s; reverse and handbrake settings are unchanged. In host Chrome, holding W for 3.2 seconds then A for 1.1 seconds changed the heading by 1.24 rad with no browser errors. This makes steering more responsive, but the full-lock capture puts the car at the road edge, so this check does not establish reliable cornering or retail handling. The vehicle force model remains approximate.
Open or save the capture · Drive the packaged city · Read the driving model gaps
A fresh host Chrome/WebGL2 load reached “City scene ready” in all eight packaged regions with no failed requests, page exceptions or console errors. Recovered route counts are L4RA 832, L4RB 99, L4RC 54, L4RD 26, L4RF 33 and L4RG 44; L4RH and L4RR have no free-roam route graph and render their authored test-track and airport pavement instead. The normal turn acceleration is now 9.0 m/s² and its yaw cap 1.0 rad/s; a short drive changed the 240SX heading by about 45 degrees. The turn tune is viewer-side, and these captures do not verify retail handling or citywide placement.
Open the packaged WebGL viewer · Read the driving model gaps
The geometry-aware region audit found 96,008 coincident placements between L4RA and L4RD: 96,006 had identical render data; the two different payloads were the global sky shells. Those same shells repeat across all eight regional streams. The combined merger now retains L4RA’s explicit `LOC4DYNTEX` bindings once and skips the seven duplicate sky pairs. The combined GLB loses two nodes; its city geometry and route data are unchanged. The host view loaded with 108,111 nearby triangles, 257 textures and no browser errors. The visual change is subtle in this view; the 240SX remains flat gray, and this does not prove citywide placement.
Open full-size capture · Read the geometry-aware audit · Open the combined drive
The common pack pads its pixel data with 0x11 bytes to a 0x80 boundary. Greedy skipping advanced three bytes beyond the aligned base for `SFX_LIGHT_BEAMA` and `SFX_LIGHT_BEAMB`, scrambling the DXT3 blocks. The parser now backs over the padding run and restores the aligned payload base. Both images decode as smooth grayscale light beams. All eight regional scenes and the combined scene were rebuilt from that correction. This matches the pack layout evidence; retail GPU output for the effect blend is still unverified. The 240SX remains flat gray, and no change was made to car paint or lighting.
Open before · Open after · Read the offset findings · Open the rebuilt L4RA drive
Hiding only the MegaLight's `SFX_LIGHT_BEAMB` mesh removes the fragment. Its exact 32×256 DXT3 source texture is recovered; the remaining gap is the retail effect's draw-state and activation. The default scene still includes it while that behavior is traced.
Open full-size diagnostic · Read the source findings · Open this diagnosticThe scene reached City scene ready with 832 free-roam routes and no uncaught page exceptions. The capture confirms the car is on marked pavement and traffic is present. It also shows the unresolved silver-gray body, pale distant traffic paint, and a bright fragmented shape at the upper-right sky edge. This records the current state; it does not establish a retail visual match.
Open full-size capture · Read the capture notes · Open this scene
Host Chrome isolation captures show the bright upper-right fragment remains with the magnolia and juniper row hidden individually. Hiding the nearby condo also leaves it present. The decoded cloud panoramas look normal. The source object/material combination is still unresolved; no city meshes were removed from the default scene.
Open magnolia diagnostic · Open juniper diagnostic · Read the isolation notes
Host Chrome / WebGL2 rendered the L4RA scene with a 128×128 local cube probe. It reached City scene ready with 129 nearby textures and no browser errors. Compared with the 64×64 capture, the 240SX gains no clear body-reflection detail. The default stays at 64; the remaining appearance gap lies in the retail material and environment inputs rather than probe resolution.
Open 64×64 capture · Open 128×128 capture · Read the comparison note
The viewer now uses the recovered per-car GLOSS alpha to set the factory-finish lighting amplitude. Clean-context host captures confirm the 240SX silver-gray and 350Z orange stock pigments are selected independently. Both reached City scene ready with 129 nearby textures and no browser errors. The paint reads more clearly, but reflections still use the WebGL approximation and traffic paint remains pale.
Open 240SX capture · Open 350Z capture · Read the source linkage and limits
The host capture shows the opt-in shader using the explicit gameplay profile, live chase-camera basis, and three nearby authored type-2 fixtures. Compared with r440’s profile-only result, the 240SX is more readable, but the body still lacks convincing paint reflections and traffic cars remain pale. The game’s exact spatial-pack traversal, active material flags, and environment bindings are still missing. This nearest-fixture selection is an experiment and remains out of the default renderer. Both captures were rendered in host Chrome; XP was used only to extract static game data.
Open full-size r441 capture · Open r440 profile-only capture · Open the opt-in diagnosticThe opt-in decompiled shader now recomputes the recovered explicit gameplay profile against the live chase-camera basis instead of reusing coefficients frozen at one camera pose. This capture rendered through host Chrome on the Quadro; the XP guest was used only for static data extraction. It confirms the scene loads, and also shows the diagnostic body is still too dark while traffic remains pale. Retail dynamic light-pack selection and material state are missing, so this profile-only path stays diagnostic and does not replace the normal driving renderer.
Open full-size host capture · Open the diagnostic drive
The current packaged L4RA scene reached “City scene ready” in host Chrome with the 240SX selected. A short W hold followed by D moved the car to 11 mph and changed its heading on the marked road. The view has ambient traffic and no obvious building-over-road intersection at this start. The capture also makes remaining visual gaps plain: body paint is still a flat gray fallback, distant traffic paint is pale, and this one viewpoint does not verify the rest of the city's placements. The XP guest was not used for rendering.
Open full-size drive capture · Open the same L4RA drive
The executable's file-backed fourth-channel normalization scales this projection by about 1.0224; the RGB normalization enable is a zero-filled global. Identical 960×540 host Chrome views show little visible difference, and both remain far from retail. The experiment stays opt-in. This rules out that small coefficient adjustment as the cause of the missing body shading.
Open unnormalized capture · Open file-default capture · Read the source inputs and limits
The viewer-only low-speed yaw cap is now 0.65 rad/s with a 5.5 m/s² lateral limit; higher speed still reduces yaw. The shorter right-steering input keeps the 240SX on the asphalt, while holding the turn for three seconds carries it to the road edge and the support check stops it. The first capture demonstrates the requested tighter response; the second records the remaining edge limit.
Open short-turn capture · Open long-hold capture · Try the driving view
Both packaged views reached City scene ready in Chrome with no browser errors. The normal shader uses the extracted factory 240SX color (#747270), distinct paint, glass, interior and wheel batches, and a live local city reflection probe. The decompiled shader experiment still lacks the retail light packet and is visibly too dark, so it remains diagnostic only. The XP guest is not used for rendering.
Open normal capture · Open experiment capture
The diagnostic now applies the statically recovered gameplay profile direction transform through the current chase-camera basis. The car is upright and traffic is present, but the 240SX still reads as flat and the overall lighting does not match the original. This is an opt-in experiment; the normal driving shader remains unchanged, and the XP VM is used only for extraction because its virtual GPU cannot render the game.
Open diagnostic capture · Read the conditional profile inputs · Open normal driving view
Static executable tracing now proves the gameplay actor renderer passes profile 0x7f3f00; the older 0x7f3e10 field belongs to frontend preview wrappers. The opt-in shader now uses the selected factory METPAINTSILVER effect/GLOSS and a conditional CPU projection of that gameplay profile. Chrome reached City scene ready without errors. The body remains too dark because the offline reference still uses identity experiment bases and cannot reproduce loaded light-pack traversal, dynamic lights, or runtime normalization. The normal driving path is unchanged.
Open diagnostic capture · Read the static call-path evidence and limits · Open normal driving view
The XP files statically identify 240SX RGB/GLOSS and its METPAINTSILVER uniform set. The exporter now joins those source inputs per car; the opt-in decompiled shader path uses the factory effect and GLOSS alpha while leaving the normal driving shader unchanged. Host Chrome reached City scene ready with no page or console errors. The experimental body is visibly too dark, so this is not promoted as a rendering fix. The player’s live lighting harmonics, material flags and technique remain unconfirmed; XP is not used as a runtime graphics reference.
Open normal capture · Open diagnostic capture · Read source linkage and limitations · Open normal driving view
The 240SX definition’s primary paint descriptor resolves to RGB 116/114/112, GLOSS 128 and METPAINTSILVER. The viewer now uses that factory tint when no saved paint customization exists; saved choices still override it. Packaged Chrome reports the source setup on the paint batches and reached City scene ready without page errors. The screenshot shows the resulting tint, while the body still looks flat: the retail effect technique, lighting selection, normalization and cube bindings are not yet reproduced.
Open or save this host capture · Read the executable/data trace and limits · Open this packaged view
The source hash identifies the 350Z badge/detail atlas as DECAL. Its six small material slices now retain their texture and alpha while avoiding the body-paint clearcoat path. A focused rebuild changed only this material name; geometry, normals, UVs and embedded images are byte-identical. Both host Chrome captures reached City scene ready on the NVIDIA Quadro P2000 with 170 city textures and a ready reflection cube. The visual difference is subtle at this distance, and the flat body paint remains unresolved: the badge image cannot replace the missing retail body material state.
Open before · Open after · Read the source comparison and remaining inputs · Open the same 350Z view
This frame came from the already-loaded L4RA viewer on host Chrome. The WebGL renderer string reports ANGLE on the NVIDIA Quadro P2000; the XP guest was not used. Road markings and traffic render, while the 350Z paint still appears flat and the traffic cars are mostly white. An earlier fresh browser capture was still uploading textures, so it was not treated as a load failure.
Open or save this full-size capture
A 2 m scan of all 832 L4RA routes checked 258,833 car-width positions where the route center had road support. Fifteen positions touched two placed groups at an industrial junction and a Harbor route edge. Host Chrome screenshots show the junction and highway. The renderer’s broad triangle-clipping experiment removed 1,117 building triangles and cut a foreground object into a wedge, so the normal scene remains unchanged. This scan is local to recovered route corridors and does not certify the whole city. Read the pairwise stream and corridor audit.
Open baseline · Open clipping diagnostic
The combined export loaded all eight source regions and 1,088 recovered routes in host Chrome/WebGL2. On an L4RB drive page, clicking a map road selected route L4RB:1, checked its lane against the combined road surface, and displayed the destination marker. Route snapping stays within the selected stream because the retail executable loads one regional stream and the recovered graph has no cross-export continuation. This is destination marking, not automatic navigation. The whole-world union remains experimental pending citywide overlap and transition checks. Read the region-scoped selection and limitations.
Open the combined drive capture · Open the L4RB map capture
The XP guest is a read-only source because its virtual GPU cannot run the retail game. Host Chrome/WebGL2 rendered the drivable L4RA scene and the new full map. The viewer reached City scene ready with 89,205 nearby triangles, 223 nearby textures, 832 routes, and 12 ambient vehicle body models. Click the map or press M while driving L4RA to open it; map clicks use the existing nearest-route destination logic. The alignment is approximate and the wider reconstruction still has known visual and traffic gaps. Read the source and alignment notes.
Open or save the map · Open or save the drive capture
The XP guest’s virtual GPU cannot run the retail game, so it was used only as a read-only source. Host Chrome/WebGL2 rendered both images on the Quadro. Correcting the reflected-vector input, diffuse alpha, and extracted environment range did not produce convincing paint: the opt-in result is much darker than the default. The experiment remains diagnostic-only pending the live retail paint binding and material state. Both captures loaded without browser errors or failed requests. Read what changed and what is still missing.
Default capture · Experimental capture
The read-only XP disk supplied all 29 playable car geometry/texture pairs. Rebuilding from those sources reproduced the existing GLBs byte-for-byte for 28 cars. The Lancer gained the shared CHROME atlas for its stock chrome material; its body paint remains flat because the retail paint state is runtime-driven. Host Chrome loaded the city and drove this car to 11 mph with no browser errors or failed requests. This is a WebGL reconstruction capture, not retail-game footage. Read the source extraction and texture audit.
The viewer now loads the selected car’s extracted AEMS loops alongside its stock Ginsu acceleration/coast pair, and uses the executable’s per-car channel mix points, volume fields, and RPM-change thresholds. Host Chrome loaded all seven 240SX AEMS streams with AudioWorklet active. A first all-car pass found and fixed a leading-zero path bug for four banks; the corrected packaged viewer reached the AEMS/Ginsu-ready state for all 29 selectable cars with no failed requests or browser errors. RPM-band selection remains approximate, and the screenshot itself contains no audio. Read the mixer trace and limits · Read the all-car bank check.
Open or save the capture
The packaged dropdown contains all 29 stock car keys. I selected each sequentially and waited for the rendered-car debug key to match; every key matched, with no failed requests, HTTP errors, page exceptions, or console errors. This checks model selection and loading, not the visual quality or orientation of every car, and it does not validate every engine sound. Read the capture scope.
Open the Audi TT capture
Host Chrome debug data confirms that all four 240SX body batches use the saved fallback color, have no body base-color texture, and do have both city and local environment textures available. Changing default finish to gloss changes the finish mode but makes little visible difference. No shader change was promoted from this check; the unresolved part is the original runtime paint binding/state. Read the live material audit.
Open the gloss capture
In the packaged L4RA build, a 3.2-second W hold followed by a one-second A hold brought the 240SX to 25 mph. The game selected first gear, displayed 4,685 rpm, played the original engine recording, and rendered four ambient traffic vehicles. The runtime reported no road failure, building or traffic collision, page exception, or console error. The car still has unresolved flat gray paint; this capture does not establish a match for retail handling or citywide placement.
Open or save the full-size capture · Open the packaged city viewerThis capture is from the current server build after the loading screen reached ready. Browser inspection found 223/223 nearby textures uploaded, no failed requests, and no page or console errors. The 240SX is on the road and traffic is visible ahead. The body still reads as flat pale gray, so the car appearance is not complete. The XP VM was not used to render this view.
Open or save the full-size capture · Open the packaged city viewer
Both captures use host Chrome/WebGL2 with the XP guest out of the render path. The experimental URL path ran without browser errors, but its result is substantially darker than the default and still does not produce convincing paint. Its harmonic inputs are conditional offline L4RA fallback coefficients at the spawn, and it uses the DEFAULT paint state because the live player material selection is not captured. This comparison shows the decompiled equations are not enough to reconstruct the car appearance without the actual runtime state and paint texture. The experiment remains opt-in at this URL; the normal renderer is unchanged.
Default screenshot · Experimental screenshot · Read this experiment's limits · Read shader analysis · Read paint-state audit
Fresh 960×540 captures from host Chrome/WebGL2 after the loading fade. The diagnostic found zero placed XB_ building groups with a 3D triangle crossing the road within 12 m of each selected start. This only checks each start area. The images also show why the region selector is not a complete Bayview map: garage, plaza, airport, and isolated-track contexts are mixed with city-road scenes. XP guest graphics were not used.
The packaged page reached “City scene ready” with 89,205 nearby triangles, 223/223 uploaded textures, 832 L4RA routes and 12 ambient vehicle models. A short W hold moved the player car about 4.3 m; the browser reported no page exceptions or console errors. The road markings come from the source road materials. This frame still shows flat gray body paint and dark glass, and it is one start point rather than proof of citywide placement. The XP guest was not used for rendering.
Open or save the full-size capture · Open this packaged drive
The current clean GLB stores the dynamic sky under source meshes SKYDOME_0_0/SKYDOME_0_1, so the old name-based lookup missed it. The viewer now binds the 512×128 TRACKS_2414a01e cloud panorama from LOC4DYNTEX.BIN; Chrome confirms texture 781 loaded and keeps it resident while driving. The normal view still has pale flat paint and dark windows. The forced reflection image is a shader diagnostic, not a claimed retail match.
A fresh host Chrome capture reached “City scene ready” after uploading all 223 nearby city textures. A brief W hold moved the 240SX to 11 mph in first gear; ambient traffic and source road markings are visible, with no page exceptions or console errors. The XP guest GPU was not used. The body still looks pale and flat, the windows remain dark, and this single L4RA view does not prove complete city coverage or retail handling.
Open or save the full-size capture · Read why the body paint is unresolved · Open this drive view
All eight region selections reached “City scene ready” in Chrome with no JavaScript errors. The captures show why that is only a load check: L4RA and L4RF have street-like starts, L4RC looks like an enclosed garage area, while L4RG is sparse and L4RH/L4RR have no ambient routes and little or no city context. L4RB and L4RD show elevated structures that still need route-by-route placement review. The sidecar filter removes event-only scenery; it does not make these streams adjoining map tiles or validate their start locations. The XP GPU is not involved in any of these captures.
Open or save the contact sheet · Read the placement-group audit · Read the world-entry track-ID cross-check
The matching retail companion bundle identifies 5,697 placements that belong only to numeric BARRIERS_* or PLAYER_BARRIERS_* groups. Those rows are now omitted from the L4RA free-roam export; shared and free-roam groups are retained. Chrome loaded the rebuilt scene with all nearby textures and no page errors. This start view looks coherent, but it does not show the filtered barrier locations or verify other streets. The 240SX body still looks flat and the recreation remains unfinished.
This is a screenshot of the WebGL recreation running in Linux Chrome; the XP guest was not used to render it. The city is loaded with road markings, roadside structures, and traffic visible ahead. The capture also makes the current shortcomings clear: the player car’s body and windows still read as flat, traffic is distant, and the road’s color and lighting remain unlike the retail game. It is one start viewpoint, not proof of full city placement or driving fidelity.
Open or save full-size capture
Linux Chrome reached a ready 64-pixel local reflection probe. Doubling it to 128 pixels, then temporarily raising the local reflection weight, did not make the body read materially more like a painted car from this view. Both shader trials were reverted. This points the next car-fidelity work toward the missing retail paint and lighting inputs, rather than probe resolution. These are WebGL diagnostics, not retail screenshots.
Open the 128-pixel probe capture · Open the reflection-weight capture
The L4RA-only start reports zero building/road triangle intersections in its local scene. The eight-region candidate reports six building groups crossing authored road triangles, and this same start gains an overhead structure absent from L4RA alone. Omitting the six groups removes that structure, but the filter can also remove valid architecture. A debug query now records exact triangle pairs with ?debugRoadIntersectionSamples; it is diagnostic evidence, not a safe cleanup rule. The combined candidate stays optional while regional scene membership is resolved.
The player-contact distance check remains separate from rendering. The previous render cull hid cars at the same 4.2 m center radius that stops the player, creating an invisible obstacle. The cull is removed. Chrome loaded the city and rendered six ambient cars without page errors; this capture does not sample a close-contact collision.
Open or save full-size capture
Chrome loaded 89,205 nearby triangles, 223 nearby textures, and 12 ambient body models. A 4-second W input moved the 240SX to 23 mph in first gear with the assigned engine recording active; the car remained on the paved road in this straight sample. The capture is an unedited browser render, not a retail-game comparison. The body paint still looks flat and distant traffic still appears pale.
Open or save full-size capture
After about 11 seconds of W with no steering input, the car reached 62 mph in third at 4,976 rpm and remained on the pavement; the roadway curves ahead, so this capture stops before that turn. A separate run sampled 61 mph at 6,396 rpm in second before the 3rd-gear shift. The small run-to-run timing difference makes this a browser control/gearbox check, not a retail handling comparison. The 130 mph cap was not reached or verified.
Open or save full-size capture
A projected scene-object check and a one-slice hide confirm the patch is `XO_MEGALIGHTA_1A_00_0_2`, using the recovered `SFX_LIGHT_BEAMB` texture key `0x2e95ce7d`. The shared pack marks it usage 2 / blend 2, and the viewer currently renders blend 2 additively. The source pixels themselves contain the saturated colors seen in the drive view. Hiding the slice removes the patch but also removes this authored light beam; I have not hidden it in the normal scene because the guest cannot provide a retail GPU comparison. This is not evidence of a condo placement error.
Open the hidden-slice capture · Open the source texture · Read the shared-pack extraction note
Both views use the same packaged 240SX drive start. Isolating the local environment makes the body very dark; isolating the sky panorama produces broad warm streaks. Raising the local reflection multiplier from 0.45 to 1.20 changed the body by less than one average RGB code over the car crop, so I did not keep that experiment. This confirms the current environment inputs cannot supply a convincing retail paint finish on their own; the game's selected paint state, bound environment and harmonic-light data are still missing. These diagnostic modes do not alter the default viewer.
Open local-reflection capture · Open sky-reflection capture · Open the normal drive viewThe exported trunk pane had been renamed after its shared REAR_DEFROSTER texture, causing the viewer to miss the original WINDSHIELD material role. The viewer now recognizes that exact source hash, applies its existing glass response, and keeps the authored defroster texture and blend mode. I reverted the unrelated global opacity experiment. The pane is brighter and reflective, but still does not match retail tint; the body remains flat gray and the distant facade artifact remains visible.
Open or save full-size capture · Compare with the previous capture · Open the same L4RA start view
This is the scene after all 223 nearby textures uploaded. The marked roadway is clear at this viewpoint and the visible traffic is upright. The 240SX body still reads as flat gray, its rear pane is too dark, and a distant facade shows a distracting color artifact. One viewpoint does not prove citywide building placement or traffic correctness, and this is not a finished game export.
Open or save full-size capture · Open the same L4RA start view
A brief W+A input moved the Skyline along the curved road; it ended at 11 mph with the original Skyline engine recording active. Chrome reported no page errors. This checks the packaged drive still runs after the steering and material changes. It does not validate retail handling or show the car under load at speed.
Open or save full-size capture
Eight source-identified CARSKIN primitives across the 240SX, Corolla, RX7 and Skyline now use the body-paint role instead of the enclosing assembly’s lamp fallback. The Skyline shows a gray rear surround with all four red lenses intact. The capture verifies this material-role correction; the body finish and pale rear glass still do not match the retail game.
Open or save full-size capture · Read the source and material note
All 29 stock cars now classify the shared WINDOWMASK material hash as glass; 82 active body pieces use that glass material, while separately named rear defrosters remain separate. This 350Z capture confirms it loads in the packaged viewer. The windows still appear dark and the body is flat gray, so matching the retail glass blend and car paint remains unfinished.
Open or save full-size capture · Read the fleet material note
The route data’s corridor adjustment is signed; the viewer now checks negative offsets as well as positive ones while retaining pavement and road-texture checks. Browser telemetry found a supported −5.740 m candidate at this route start. The cars shown here selected zero offset, so this image checks scene loading and vehicle orientation but does not show the changed lane candidate.
Open or save full-size capture · Read the traffic note
The center and lane markings are visible, and no building crosses the paved roadway in this frame. Retaining walls and storefronts line the sides; several ambient cars are ahead. This local view does not establish citywide placement or retail traffic behavior.
Open or save full-size capture · Read the viewpoint note
The bright diagonal across the rear glass was the authored rear wiper, using a missing DULLPLASTIC texture. Only that identified primitive now receives a dark neutral trim fallback. The rear pane and source-textured defroster are unchanged. The exact retail wiper material remains unresolved.
Open or save full-size capture · Read the material note
The three active stock body slices using retail hash 0x3ed70c43 now enter the viewer’s glass path instead of opaque paint. The separate source-textured REAR_DEFROSTER overlay remains visible. In this capture the rear and side glass still read very dark; this only fixes the material role and retains the viewer’s 0.76 glass-opacity preview, not the retail blend state.
The combined scene now loads each region’s recovered ambient routes, with traffic state keyed by region and route ID. Chrome loaded 1,088 filtered routes across L4RA, L4RB, L4RC, L4RD, L4RF and L4RG, and remained ready at an L4RB route start with all 229 nearby textures uploaded. Three ambient cars were rendered ahead/nearby in the checked frame. Transition records are matched only inside their source region; no cross-region joins are inferred. Player start selection and density polygons remain selected-region-specific. An offline 5 m route sampling audit found zero cross-region route pairs with more than 50% mutual XY coverage within 0.8 m and aligned tangents. This checks that the route union runs on the packaged combined scene; the traffic preview still does not reproduce retail traffic control.
Open full-size L4RB capture · Open combined L4RB drive · Read route counts, dedupe rule, and limits
Older capture. Its building rotations are wrong and it is kept here as evidence of the defect being corrected.
Open or save full-size capture
Current inspection capture of the assembled body, tail lights, glass, and placed wheels. The body geometry remains visibly coarse.
Open or save full-size capture
Current front inspection view showing headlights and the placed front wheels.
Open or save full-size capture
Older capture with the same transposed scenery rotations as L4RA; the currently served GLB has the corrected placement matrices.
Open or save full-size capture
Top-down diagnostic from the corrected GLB, showing 17,000 pavement triangles and building triangles within one metre of the drive-start elevation. The selected start sits on the traffic route centerline; nearest building bound is 16.64 m away.
Open or save full-size audit
Earlier 1280 × 720 capture after changing the scenery transforms. The screenshot did not verify full scene placement; the current review below records that problem as unresolved.
Open or save full-size capture
Fresh 1280 × 720 capture of the live viewer. The stored placement bounds match the corrected transforms, but the rendered road and structures still do not read as a clean, correctly assembled street scene. This is not a completion image.
Open or save full-size capture
Triangle-level top-down projection at the current start. Red means a building triangle projects over road space at some height; this is not proof of a ground-level collision. A separate height-band sample found no XB surface crossing the road between pavement height and 2.2 m above it within this local crop. The elevated projections explain the imposing overhead structure in the driving view, but their intended retail placement is still unverified.
Open or save full-size audit · Read the placement audit
Fresh SwiftShader capture after filtering the authored XO_TRACKBARRIER_PLAYER_00 meshes. This removes 18,720 nearby triangles. A second closure family remained farther along the street and is shown in the next comparison.
Quadro backed Chrome capture after driving along the selected route. The blank panel wall in the earlier view was authored XO_TRACKBARRIERB_A_1A_00 geometry intersecting the road, not a casino building. It is now filtered along with XO_TRACKBARRIER_PLAYER_00 in free-roam mode. The Keno/Craps casino storefront and overhead arcade are also present in this retail-game screenshot. This fixes the identified blocker; full city placement and lighting still need broader review.
Fresh 1280 × 720 SwiftShader capture after the name-safe duplicate cleanup and suppression of PAN_HOTELS and PAN_ENTERTAINMENT in the background pass. The distant silhouette clutter is reduced, but the broad road and overall city composition still look wrong. This is an in-progress diagnostic, not a completion image.
1280 × 720 capture of the latest cleaned scene, at route 450 position (−505.68, −513.28). The road texture now renders opaque. The casino overhang is 5–6 m above pavement, matching the kind of arcade visible in the retail reference; this capture does not prove all surrounding building placements or LODs are correct. The buildings still need targeted review, so this is not a completion image.
Open or save full-size capture · Read the placement audit
1280 × 720 view after removing the second ambient dimming pass from the game's baked vertex colors. Façade and arcade details read more clearly, while the lane surface remains too dark; dynamic light behavior still needs calibration. The capture helps inspect placement but does not verify the rest of Bayview.
Open or save full-size capture
Fresh 1280 × 720 capture at route 1055, distance 78. The car is passing through the authored XB_SHORTCUTTUNNEL area; this explains the metal enclosure at this spot but does not account for the reported misplaced city buildings. Open this exact view.
Fresh capture after the exporter began reading asset names from the fixed source-header field. The scene pixels in this casino view are effectively unchanged, so that correction does not explain the buildings the user reports as misplaced. It does correct panorama and source-family classification elsewhere in L4RA. Open a nearby route at the corrected hotel panorama. This is not completion evidence.
Open or save full-size capture
Fresh capture at route 534, distance 72.3, with a 200-unit scene radius for this diagnostic. The source panorama’s ground triangles crossed the authored road at car height; it is now excluded from the nearby background pass. Nearby hotel buildings and road remain visible. Open this exact view.
Open or save full-size capture
Older SwiftShader capture at L4RA radius 120. The road lines shown here came from a prior generated-marking experiment and are not representative of the current viewer. Current road paint samples the decoded RDP textures and UVs directly; the asphalt detail and car glass still need work. The 240SX engine recordings were extracted later; see capture 19.
Open or save the older capture · Open the current scene
The selector switches to the extracted Coupe traffic body and keeps it on the road. Its body-only extraction is visibly rougher than the 240SX, so this is functional selection evidence, not a finished garage roster.
Open or save full-size capture
The Ginsu recordings were recovered from the XP install. Chrome decodes the original EA-XAS recording, switches between acceleration and coast samples, and updates the RPM loop while driving. This screenshot records the active sound state; it cannot demonstrate speaker output. Other selected bodies still need their matching player audio mapped.
Open or save full-size capture · Open the scene and press W
Fresh 1280 × 720 browser capture after replacing the guessed traffic axis mapping with the recovered X-forward, Y-up, Z-lateral basis, and removing incomplete traffic shells from the player-car menu. Only the exported 240SX is currently a player vehicle. Its visible wheel placement and the remaining city presentation still need work; this is not completion evidence.
Open or save full-size capture · Open this scene
Fresh 1280 × 720 SwiftShader capture after loading the 3000GT from its default KIT00 parts in the XP install. The selector now lists all 29 retail player-car folders; traffic-only shells are excluded. The tire locations are still approximated from model bounds and the car body remains too flat, so the model needs more visual work. The Vulkan capture attempt lost its WebGL context during city texture upload on this machine. Only the 240SX has its original engine recording wired. This does not claim the full city reconstruction is complete.
Open or save full-size capture · Drive this 3000GT view
Fresh 1280 × 720 SwiftShader capture. The normal traffic roster now skips zero-quota assets, reverses triangles for the three bodies with reflected source transforms, and begins route motion after the traffic meshes finish loading. The cars in this chase view remain distant, so this capture does not prove every model's direction or assembly. Traffic still lacks its retail tire attachment transforms.
Open or save full-size capture · Open the current drive view
This is the latest 1280 × 720 SwiftShader capture with the 3000GT selected. All 15 traffic exports now include their authored BODY_A and TIRE_FRONT_A geometry, and the viewer estimates four tire positions from body bounds. Traffic is too far away in this view to verify wheel placement, orientation, or direction. The player car remains visibly flat and the road presentation still needs work; this is diagnostic evidence, not a finished result.
Open or save full-size capture · Open the current drive view
Fresh Chrome capture with the 3000GT selected. The car shader now adds view-dependent clear coat and blue glass reflections, and the same finish path is applied to traffic. The windows read more clearly, but the body is still too flat and its surface detail is low compared with the original game. This does not validate traffic orientation or wheel positions.
Open or save full-size capture · Open this drive view
The garage manifest now carries all four stock axle/track positions from GLOBALB.BUN for every selectable car. This drive check selected the 3000GT and pressed W: its speed rose above zero, while the HUD correctly states that no 3000GT recording is mapped. The previous build played the unrelated 240SX sound here; that cross-car playback is now blocked. Close visual confirmation of the wheel fit is still pending.
Open or save full-size capture · Open the 3000GT
I cycled all 29 stock car entries in Chrome, held W for each selection, and observed the HUD leave zero speed. Every selection reached “City scene ready”; the browser recorded no page or console errors. This capture shows the last selection, an Audi TT. The available cars are 240SX, 3000GT, 350Z, A3, Celica, Civic, Corolla, Eclipse, Escalade, Focus, G35, Golf, GTO, Hummer, Impreza WRX, IS300, Lancer Evo VIII, Miata, Mustang GT, Navigator, Peugeot, RSX, RX-7, RX-8, Sentra, Skyline, Supra, Tiburon and TT. This verifies loading and movement, not that every model has a finished visual fit.
Open or save full-size capture · Open the Audi TT drive view
Fresh 1280 × 720 Vulkan Chrome capture from route 450. This ambient car is upright and faces away along the route. Traffic now travels to the end of its recovered directed path instead of teleporting to the first waypoint and snapping to a new heading. This checks one model on one route; tires and body detail remain rough, and the full traffic fleet still needs close inspection.
Open or save full-size capture · Open this route view
The Windows XP install contains named Ginsu recordings for a subset of the garage. I mapped 13 cars using recordings named for that car or its exact model family; the G35 is loaded here and its HUD follows the revs while W is held. Sixteen cars have no safe named match and stay silent. That includes the 3000GT, which no longer borrows the 240SX recording.
Open or save full-size capture · Open the G35 drive view
Current Vulkan Chrome capture after route-end traversal changes. The two visible traffic cars stay upright on this sample route, and the player can select the 3000GT. Their bodies remain pale and crude, and the surrounding city lighting and road surface still fall short of the game. The full 29-car roster loads and moves, but that does not mean its visuals are finished.
Open or save full-size capture · Open this drive view
The Impreza has no wheel object in its selected geometry export. Its fallback now retains the Mustang wheel texture atlas, so tire and rim surfaces show up instead of plain white disks. The rim is generic and the fender geometry still looks coarse; this is a verified parts assembly improvement, not a finished car.
Open or save full-size capture · Compare Lancer Evo VIII
The same fallback is placed at this car's recovered axle positions. It restores visible tires and rims, though the white wheel arch pieces and body surface still need correction.
Open or save full-size capture · Inspect the stock Lancer
Chrome loaded the selected Impreza, advanced to 24 mph under throttle, and showed its named original engine recording. The fallback wheel textures load without JavaScript or WebGL errors. This rear view verifies selection and movement; it does not certify wheel fit or city fidelity.
Open or save full-size capture · Drive the Impreza
The old stripe followed the traffic car path, so it ran under the player car. This capture shows the revised estimate: where opposing route paths are close and face each other, the stripe sits midway between them and snaps to the pavement triangles. It is an inferred lane divider; roads without a matched opposing path rely on their source material and still need review.
Open or save full-size capture · Open this road view
The traffic lead was measured in seconds while the viewer clock uses milliseconds. That unit mismatch put candidate cars at the player, where the overlap guard removed them. After fixing the conversion, a fresh Chrome capture shows a car ahead on route 450 with the 3000GT clear. This verifies visibility at one stop; the traffic body detail and complete fleet behavior still need review.
Open or save full-size capture · Open this drive view
A fresh browser sweep selected all 29 stock retail entries. Every selection returned to “City scene ready”; Chrome reported no page or console errors. The final selection is the Audi TT shown here. This checks the selector and asset loading; it does not verify each car's visual fit or driving feel.
Open or save full-size capture · Open the Audi TT
The exporter had dropped the source vertex normals, so the shader treated every car triangle as a flat face. This regenerated 3000GT inspection capture uses the retail normals for its curved paint and glass. Chrome reached “3000GT stock model ready” with no browser errors. The body still has visible geometry limits and needs work against an in-game reference.
Open or save full-size capture · Inspect the 3000GT
The current 3000GT body and glass use the same recovered normals in the road scene; one ambient car is visible ahead without overlapping the player. The 3000GT stayed at zero speed for this capture, so this checks materials and traffic placement rather than driving motion. Road texture and city placement remain visibly short of the retail game.
Open or save full-size capture · Open this drive view
A close route check showed the exported traffic bodies' nose points along local −X. The viewer had mapped local +X to the route tangent, so traffic moved with its front facing the player. Mapping local −X along the route now shows the rear as expected. This capture checks one sedan on one route; traffic body textures and wheel placement still need work.
Open or save full-size capture · Open this route view
The 29-option menu switched the live car to the stock Acura RSX. Holding W moved it to 5 mph with no browser errors. This verifies a live car change and motion; it does not establish polished handling for every car.
Open or save full-size capture · Drive the Acura RSX
Each normal-quota ambient body was placed at the same route pose with the corrected local forward axis. The close sweep shows all twelve upright on the pavement; the body materials and estimated wheel positions remain visibly rough. This checks orientation across the model roster, not every route or traffic transition.
Open or save full-size lineup
After fixing the frame-time cap, the served page reached 42 mph while W was held for six seconds, with no browser errors. This capture was taken just after releasing W while coasting at 16 mph; the car remains on the pavement and an ambient car is ahead. This checks one car on one route. The city still looks dark and the road marking is missing in this stretch.
Open or save full-size capture · Open this drive view
The recovered road triangles show a stable, roughly 20 m pavement band here, but traffic data has only one direction. The viewer now infers a center path from that band and renders short yellow dashes. On the served page, the 3000GT reached 52 mph during a six-second throttle hold with no browser errors; this capture shows it coasting at 20 mph with the markings and traffic visible. The inferred line still needs comparison with original gameplay, and the road lighting remains too dark.
Open or save full-size capture · Open this drive view
Chrome was premultiplying the RGB of low-alpha road overlays before the shader composited them, effectively darkening those colors twice. The texture upload now disables that conversion. This served-page capture shows the road and lane marks brighter than the previous capture; it still needs comparison against retail gameplay, and the skyline remains unresolved.
Open or save full-size capture · Open this drive view
The traffic exporter retains raw game coordinates: +X is forward, +Y is lateral, and +Z is up. The viewer had treated Y as up and mapped −X forward, which caused the odd orientation and reverse-facing cars. This capture follows the corrected basis on route 450; traffic body color and detail still need work. Chrome also confirmed that all 29 stock player cars appear in the selector.
Open or save full-size capture · Open this drive view
Fresh Chrome capture after holding W and D for six seconds, starting from the diagnostic override at route 450, distance 7.2. The 3000GT left the pavement and reached the retaining wall, then slowed almost to a stop. This start is wall-adjacent and is a poor handling benchmark; city loading, car selection, the numeric RPM HUD, and pavement render. The 130 mph cap was not reached in this maneuver.
Open or save full-size capture · Repeat the route check
Fresh Chrome capture from the normal start, showing the assembled city, pavement markings, traffic prop, selected 3000GT, and numeric RPM. The browser read 36 mph in first gear after a five-second W hold; by the screenshot, the car had slowed to 4 mph as the route curved and the straight heading ran out of supported pavement. This confirms movement, not stable cornering or the 130 mph limit.
Open or save full-size capture · Open the normal drive start
Fresh 1280 × 720 capture with the requested Vulkan/ANGLE flags. The WebGL2 city loaded with no page or console errors while GPU usage was idle; the 3000GT reached 49 mph in 2nd at the state read, then was captured coasting at 42 mph and 4,189 rpm. The 2nd-gear calibration is 100 rpm per mph, so 60 mph maps to 6,000 rpm. Earlier context loss coincided with 99% GPU utilization on the shared card; entry 48 verifies the 130 mph cap separately. The skyline and car finish still need work.
Open or save full-size capture · Open this Vulkan tested scene
SwiftShader Chrome capture on recovered route 1013. Holding W on its long straight reached the configured 130 mph cap, 6th gear, and 6,760 rpm with no page errors. This confirms the revised six-speed mapping and speed limit. The screenshot also shows the nearby-city stream uploading textures and the incomplete horizon at this location; it is drivetrain evidence, not a finished city view.
Open or save full-size capture · Open the tested straight
After the streamer change, the replacement road and buildings stay hidden until the required textures finish uploading. The ready-state capture still reaches 130 mph in 6th at 6,760 rpm with no page errors, but this route reveals large black gaps beyond the pavement and fragmented distant scenery. Atomic swapping fixes untextured transition frames; it does not fix this location's scene coverage or horizon.
Open or save full-size capture · Open the tested straight
This capture came from a real-time Chrome session after the city and its nearby textures finished loading. The page reported “City scene ready,” 32,895 nearby triangles, and 165 nearby textures with no console errors. The view has a visible pavement surface, lane markings, storefronts, traffic assets, and the 240SX. A runtime DOM check found all 29 stock-car choices. This is a stationary start view; it does not verify driving quality or complete city coverage.
Open or save full-size capture · Open the captured start
On recovered route 1013, a live Chrome control reached 130 mph in 6th at 6,760 rpm with no console errors. This screenshot was captured just after releasing throttle, at 117 mph and 6,078 rpm. It confirms the cap in the served build. It also exposes an unresolved visual defect at this location: dark, incomplete highway scenery and weak road markings.
Open or save full-size capture · Open the tested straight
The brown rectangle was a retail ZPM start/finish graphic leaking into the free-roam scene. A road-only render and prefix-isolated render traced it to those event meshes; hiding them removes the rectangle. This fresh served-page capture shows the 3000GT on the continuous pavement at route 1013, with 53,897 nearby triangles, 194 textures, and no browser errors. The other highway scenery and lighting still need work.
Open or save full-size capture · Open the corrected route
Fresh capture from the live 8095 server after the 130 mph speed-cap cache refresh. It confirms the page loads and the race decal is gone, but this highway view still has weak lane definition, visible yellow texture bands, sparse traffic, and a dark incomplete horizon. The 3000GT is stationary in this frame; this is a defect record, not a completion image.
Open or save full-size capture · Open this exact view
After reducing the prototype throttle acceleration, a live Chrome run rose from 13 to 74 mph over six seconds instead of reaching 130 mph in four. This capture followed the run at 49 mph while coasting and shows the car moving around the fountain plaza. The pavement here is broad and poorly marked, so this confirms movement but not reliable street following; ambient traffic and city coverage still need review.
Open or save full-size capture · Open the default start
The previous default began about 85 route units before its fragment ended. The start selector now requires at least 180 units ahead; the live build selected L4RA route 716 with about 274 units remaining. This view has storefronts, a marked street, and loaded city geometry. It verifies the start selection, not full-route continuity or correct traffic models.
Open or save full-size capture · Open the updated default
The synthesized center stripe was reduced from 0.30 m to 0.15 m total width and shifted to a warmer yellow. The live r12 capture keeps the line readable on the storefront street without the former oversized stripe. Road texture bands and driving through route turns remain unresolved.
Open or save full-size capture · Open this current start
The traffic GLBs did not contain runtime paint colors, leaving every body the same gray fallback. The viewer now applies a stable color per route to the recovered paint material while keeping the separate window and tire textures. This fresh r13 capture shows the change on an SUV; the low-poly source body and exact retail color selection still need improvement.
Open or save full-size capture · Open the current traffic view
Live r14 Chrome run on the selected city start: a short left input followed by throttle kept the car moving along the recovered bend, reaching 52 mph in 2nd gear. This is more controllable than the earlier full-lock spin. The car is close to storefront geometry in this frame, so road/building placement and longer-route continuity remain unresolved.
Open or save full-size capture · Open the current start
The viewer now uses exact-duplicate-cleaned GLBs for every region. A fresh L4RD Chrome load reached “City scene ready” with 22,683 nearby triangles, 108 textures, 26 routes, and no browser exceptions. The screenshot also shows the remaining visual work: this industrial road is unusually broad, lacks a readable center marking, and has incomplete distant scenery.
Open or save full-size capture · Open this region
Fresh Vulkan Chrome capture after limiting the asphalt composite to lane, intersection, and parking materials. The recovered center line is now visible in this L4RD view and the page loads without exceptions. The surface still reads too broad and flat, so source pavement widths, material appearance, and distant city coverage remain under review.
Open or save full-size capture · Open this region
The asphalt overlays now blend over the retail vertex colors instead of a flat gray. Chrome reports the L4RD scene ready with 22,683 nearby triangles, 108 textures, 26 routes, and no page exceptions. Pavement detail is easier to see, but the lamp pool clips to white and the road still looks too wide and plain.
Open or save full-size capture · Open this region
Reducing the simulated headlight beam did not remove the large white patch. That points to the nearby city fixture light as the source; the shot also shows the recovered yellow center marking curving away from the stationary car. Both lighting balance and line alignment still need work.
Open or save full-size capture · Open this region
A second headlight reduction leaves the hotspot nearly unchanged, confirming that the remaining excess comes from the city lighting. This diagnostic capture keeps the still-unresolved broad road, sparse horizon, and vehicle-to-marking alignment visible.
Open or save full-size capture · Open this region
The fixture lights now use shorter radii and lower intensities. The lit patch is smaller, but still too bright; the asphalt remains broad and flat, and the center marking does not stay aligned with the stationary car through this bend. This is the current GPU capture, not a finished city view.
Open or save full-size capture · Open this region
The served default scene loads with 100,440 nearby triangles, 324 textures, 832 routes, and no page exceptions. The shop street is more readable than L4RD, but the road texture is coarse, the lane stripe runs under the stationary car, and the car body highlight is too bright.
Open or save full-size capture · Open the default start
Lowering the shader highlight stops the roof from clipping to white and preserves the body’s contours. The rear window remains brighter than intended, so this screenshot prompted a separate glass-reflection adjustment; the street and lane alignment defects remain.
Open or save full-size capture · Open the default start
The reflection is narrower and more blue, but the rear glass still reads as a bright panel. This requires another material/source-texture check. The capture confirms the shader change is live and the rest of the city remains visibly unfinished.
Open or save full-size capture · Open the default start
The stock garage build shows clearly tinted windows from a rear three-quarter view. That isolates the chase-view brightness to a different default vehicle export and reflection angle, rather than a missing glass material or texture.
Open or save full-size capture · Inspect the stock 240SX
This older default 240SX still uses the one-off GLB, unlike the garage inspection. Its chase view shows the bright window reflection while the stock garage model is dark and legible. This comparison led to using the same retail KIT00 GLB for both the default drive and the garage selector.
Open or save full-size capture · Open the current default
The default drive now loads the same KIT00 stock GLB as the garage selection. The rear window reads as tinted glass, the complete stock body and wheels render, and the original Nissan 240SX audio status is shown. This capture also exposes the remaining lane issue: the inferred center stripe passes through the car at the selected route point.
Open or save full-size capture · Open the default start
On route 716, the viewer detects there is no opposing traffic path and shifts the starting car three metres to the right of the inferred road center. The stripe no longer runs under the car. The adjacent shopfront and coarse pavement still need placement and material review, so this is a lane-alignment check rather than completion evidence.
Open or save full-size capture · Open the current default
Fresh Chrome capture from /dist/ on the served 8095 build. The page fetched the chunked L4RA package and reached texture upload with 32 of 324 images processed at this capture point. The loading screen is visible and updating, but startup remains too slow to call the package ready for comfortable play; this entry records the incomplete load honestly.
Fresh Chrome capture from the packaged build after the first road texture uploaded. The 240SX, pavement, center markings, drive controls, and 0 mph HUD are visible while the remaining city textures continue uploading (49/324 at capture). This shortens the time spent behind the loading screen; it does not improve the unresolved broad pavement, close storefront placement, or sparse traffic.
Open or save full-size capture · Open the packaged drive
The source road cross-section changes from its roadway surface to a side strip about four metres from route 716. The prior three-metre player offset placed the car at that edge; this build uses 1.5 m. The first playable frame shows the car beside the lane marking while the rest of the nearby textures continue loading. The pavement remains visually broad and plain, and the storefront is still close to the drive path.
Open or save full-size capture · Open the packaged drive
Captured from the packaged viewer in Chrome after a 2.2-second W hold. The 240SX accelerated from rest to 21 mph in first gear; the page had no console or runtime errors. The car remains on the roadway in this short sample. It does not establish long-route handling, and the narrow storefront clearance and rough road appearance remain visible.
Open or save full-size capture · Open the packaged drive
Captured from /dist/ in Chrome after the viewer reported “City scene ready” and all 183 nearby textures were uploaded. The early white texture placeholders are gone and the stock car is visible. This confirms the ready state renders, while also documenting coarse dark pavement, sparse lane markings, no visible moving traffic in this view, and tightly placed storefronts.
Live Chrome capture after the road support change. The 3000GT reached 130 mph in sixth at 6,760 rpm on route 1013, with no page errors. The car crossed the earlier mesh seam without the rear-footprint speed drop; it slows when the center reaches the route's actual pavement edge. This frame also shows a narrow dark road strip against a pale surface and a green traffic van rolled sideways. The speed cap works; road reconstruction and traffic orientation remain unfinished.
Open or save full-size capture · Open the tested route
The traffic body bounds place the vertical dimension on local Y for the standard car and van models, while BUS retains its separate Y-forward, Z-up frame. The source exporter and packed manifest now use those axes. In the live Chrome capture, the nearby traffic car is upright; the viewer reports 12 ambient models and has no page errors. The pale highway surface and narrow striped pavement strip remain visibly incorrect.
Open or save full-size capture · Open the 130 mph capture
The route 716 pavement probe measures a 6.5 m supported band, narrower than the old 12 m cutoff for fallback markings. The generator now accepts stable 6.5–25 m bands and places the stripe at the measured pavement midpoint; the car remains 1.5 m to the right. The fully loaded Chrome view has no page errors. Coarse cobbled road texture and close storefronts still need work.
Open or save full-size capture · Open the default drive
The extracted RDP_ROADSTRIP_COBBLEROA image has a 14% mean alpha, so it is a sparse overlay rather than a complete opaque road tile. The renderer now composites road-strip overlays over the source vertex-color base, without applying the lane-marking brightness gain to the whole pavement. The ready Chrome scene has 183 nearby textures and no page errors. It is more legible than the dark multiply pass, but the road remains warm, coarse, and unlike the target asphalt.
Fresh capture from the served /dist/webgl-city/ page using the machine’s direct headless Chrome and Vulkan flags. The page reached “City scene ready” with 100,440 nearby triangles, 324 textures, 832 routes, 12 ambient vehicle models, and no console errors. This confirms the larger default stream finishes loading. The street remains too warm and coarse, the sidewalks and shopfronts appear very close to the drive path, and the city still does not match the game’s finished nighttime look.
The direct-Chrome packaged build reached “City scene ready” with all 324 nearby textures and no console errors. I reduced the warm vertex-color cast on the road while preserving the extracted texture overlay; this frame shows gray pavement and visible yellow center markings. The surface is still unusually broad and flat, and the storefront clearance needs route-level inspection, so this is not completion evidence.
Open or save full-size capture · Open the packaged drive
The XP installation’s raw Supra vertices already use the common car axes; its per-part matrix incorrectly rotated width into height. The exporter now omits that matrix for the Supra, and all 29 stock-car GLBs were regenerated from the XP installation. In the packaged drive, the Supra accelerated to 22 mph in first gear with no page errors. Its matching original engine recording is absent, and the broad pavement and close storefronts still need work.
Open or save full-size capture · Drive the corrected Supra
Each stock-car GLB from the XP installation was opened in the WebGL inspection view in direct Chrome. All 29 reached their ready state, with no page errors. The contact sheet shows their assembled body shapes and wheels for comparison; engine recordings remain available only for the 13 models with named matching Ginsu samples.
Open or save full-size contact sheet · Open the car selector
Direct Chrome run from the former route 716 spawn. Holding throttle without steering reached the edge of its supported road surface after about 40 m, and the car appeared partly inside a storefront. Revision r47 replaced that spawn with route 403 after checking its 180 m supported runway. The screenshot remains as a record of the defect that prompted the change.
Open or save full-size capture · Open the current default start
A controlled run on route 762 covered about 114 m without a road-support failure. This is a longer usable stretch than the default start, but the roadway is broad and mostly unmarked, with a retaining wall and sparse street detail. It is useful as a drivability reference and does not resolve the city-street layout.
Open or save full-size capture · Open this route
The 3000GT route previously used for the 130 mph check also remains drivable over this longer run. The screenshot shows the export's main visual limitation clearly: the highway surface is unusually broad, with weak lane markings and incomplete background composition. The 130 mph cap is confirmed in the packaged viewer; overall scene fidelity remains unfinished.
Open or save full-size capture · Open this route
Revision r47 selects route 403 for L4RA only after checking pavement support at 20 m intervals across a 180 m runway. In a fresh packaged build, the default 240SX reached 53 mph after eight seconds of throttle without losing road support. This is a more usable start than route 716, but the road remains broad and the surrounding city composition is sparse.
Open or save full-size capture · Open the current default start
The WebGL shader now uses the retail RDP_ROADSTRIP_FREEWAYCE tile beneath sparse road and lane overlays. A fresh WebGL 2 capture shows asphalt grain and the authored road markings on the default L4RA arterial; the page reached “City scene ready” with no console errors. Road width and the surrounding building placement still need work.
L4RC has no embedded FREEWAYCE road material, so this capture verifies the extracted shared underlay is also applied there. The scene loads with no console errors and shows its road surface and ambient traffic. The nearby structure crowds the camera view, leaving the region's building placement unresolved.
This capture is from the actual /dist/webgl-city/ URL after packaging. The page reached “City scene ready,” loaded 108 nearby textures and the original 240SX engine recording, and had no browser-console errors. It verifies the distributed road texture asset loads with the packaged viewer.
Runtime capture from the all-car driving audit. The hood's stored 2.498 scale and rotation pushed it several metres away from the body.
Open full-size before capture
The exporter now omits that transform for ECLIPSE_KIT00_HOOD_A. The packaged driving view reached “City scene ready” with no console errors; the hood is seated on the car. The rest of the car and city remain under visual review.
Runtime capture from the same audit. The base object's transform stretched the body bounds and left a large dark panel beside the SUV.
Open full-size before capture
The exporter now omits the erroneous transform for ESCALADE_BASE_A. The packaged driving view reached “City scene ready” with no console errors; the detached panel is gone and the body bounds are coherent. The car still has the same unfinished materials and city limitations as the rest of this reconstruction.
Fresh capture from the packaged WebGL 2 viewer at route 450, distance 518. The arcade and storefronts frame the roadway; the page reached “City scene ready” with no console errors. The L4RA export now emits only the first valid resident mesh for each referenced solid key. A route-wide triangle audit found the only building-surface crossings at the named shortcut tunnel and warehouse drive corridors. This capture does not prove the casino view matches retail placement.
Open or save full-size capture · Open this casino street · Read the placement audit
Fresh packaged Chrome capture after strengthening the stock glass material's blue-gray Fresnel reflection and broad highlight. The windshield and rear glass now read more clearly against the silver body. The scene reached “City scene ready” with no console errors. This is a material approximation; it does not recover the retail environment reflections or glass transparency.
Open or save full-size capture · Open the packaged drive
The fixed highway spawn is removed. The selector now ranks routes near buildings and verifies the car-width pavement footprint every 10 m across a 180 m runway before choosing one. The packaged build selected L4RA route 716 near the storefronts; it reached “City scene ready” with no browser errors. A short SwiftShader throttle check stayed on the pavement, though software-rendered speed was too low to establish long-run handling. The close storefront spacing and road appearance still need review.
Open or save full-size capture · Open the current default start
Fresh capture of the served /dist/ build on the NVIDIA Quadro P2000. It reached “City scene ready” with 32,420 nearby triangles, 146 textures, and 12 ambient traffic models. The cyan cross is a debug marker. This confirms one successful load of the current scene; the broad pale road, storefront spacing, and match to retail placement remain unresolved.
The L4RA package now binds the two source LOC4DYNTEX.BIN textures to the matching SKYDOME slices. Authored cobble road and sidewalk materials also keep their own vertex-color bases instead of inheriting the freeway asphalt underlay. This SwiftShader capture reached “City scene ready” with 148 nearby textures and no browser errors. The street still has broad pale pavement, incomplete center markings, and close storefronts; a retail layout match remains unresolved. Two P2000 captures lost their WebGL context during texture upload, so this image uses SwiftShader.
The viewer now reads 2,234 light records extracted from the L4RA STREAM bundle, using their recorded positions, colors, and radii instead of estimating fixture positions from mesh bounds. Runtime intensity is still an approximation. This packaged Chrome frame reached driving available while nearby textures were still streaming. It shows that the city geometry and traffic load, and also shows the unresolved broad pale roadway, missing road detail, and storefront placement problems.
Open or save full-size capture · Open the packaged L4RA view
Fresh capture of the served package with the recovered asphalt underlay limited to lane and intersection composites. The darker lane strip and center markings read more clearly here. This frame reached driving available with 112 of 133 nearby textures uploaded, so it is an interim material check. The wide pale pavement and flat gray structures at the horizon remain unresolved; this is not a finished scene.
Open or save full-size capture · Open this route
Fresh capture after all 248 nearby textures completed uploading, at the normal 240-unit scene radius. The larger radius fills in more of the distant streetscape, so the earlier cutoff at radius 120 was a stream boundary. The broad pale pavement and several flat gray structures remain visible with every nearby texture loaded; their cause is still unresolved.
Open or save full-size capture · Open this full-radius route
The road shader now adds a restrained, broad specular response from nearby recovered street lights and the player headlights. The packaged Chrome view loaded all 248 nearby textures without browser errors. The effect adds a small highlight response; it does not resolve the broad pale pavement or the flat gray structures.
Open or save full-size capture · Open this full-radius route
Fresh capture of the packaged build: a straight-throttle run reaches exactly 130 mph with no browser errors or overshoot. The selected route curves ahead, and the car is visibly on the pale shoulder beside the dark asphalt strip. This confirms the speed setting; it does not confirm correct road placement or successful cornering.
Open or save full-size capture · Open this route test
This debug run stopped near the tunnel wall when the sampled road surface ended under the car center. Relaxing the tire-edge check did not fix missing or misaligned drivable pavement. The steering input was held straight through the tunnel; manual steering and the recovered road placement still need work.
Open or save full-size capture · Open this route test
The packaged scene applies source-opaque texture handling only to lane and asphalt-strip materials; cobblestone road and sidewalk keep their sparse-overlay compositor. The storefront view has yellow center marks and no browser errors. Its cobblestone roadway is still pale, and the sidewalk edge still reads as a large patch beside the road. This SwiftShader capture had 112 of 148 textures loaded; visual calibration and the city layout remain unresolved.
Open or save full-size capture · Open the default drive view
Packaged Chrome/SwiftShader capture after correcting traffic slot timing. The density polygon selects a two-car budget, and the live viewer reports two rendered vehicle models with no browser errors. The cyan crosses are debug route markers. Hiding the control panel reveals more of the road; the pale pavement, storefront spacing, and limited route coverage remain visible issues.
Open or save full-size capture · Open the default drive view
The shader now gives the authored cobblestone roadway a darker base while leaving its separate sidewalk material intact. This packaged Chrome capture shows the two-car traffic pack and broken yellow center marking; runtime reported no page errors. The pavement is more distinct from the sidewalks, but the street geometry, storefront spacing, debug crosses, and overall match to retail gameplay still need work.
Open or save full-size capture · Open the default drive view
The guessed total-car rule by density class is gone. The viewer now uses the recovered seven-car population cap, per-model quota weights, and a 1.5 m lane offset beside the center marking. In this default route capture, three cars render; other slots fall outside the view corridor or overlap on this single available route. The cyan crosses mark traffic positions for this debug capture. Chrome reported no page errors. The road and city layout still differ from retail gameplay.
Open or save full-size capture · Open the default drive view
Fresh capture from the served /dist/webgl-city/ package in Chrome with no debug overlays. The viewer reached “City scene ready” with 100,440 nearby triangles and 289 nearby textures; the 240SX and ambient traffic are visible on the storefront street. The roadway still has sparse markings and the lighting remains a rough approximation. A local triangle audit found no building surfaces in the default road corridor at car height, but appearance and retail placement remain unresolved.
The wheel mesh was being lifted by roughly one tire radius even though the full-car ground offset already used the tire's lowest source bound. Wheel placement now keeps native tires at that bound and aligns fallback tires to the target car's source ground height. A fresh packaged Chrome inspection view shows the rear and front tires assembled, with no page errors. This corrects the wheel height; the driving scene's road appearance and city layout remain under review.
Open or save full-size capture · Inspect the stock 240SX · Return to driving
The Impreza has no source wheel mesh, so the selector assembles the generic textured wheel set. After aligning fallback tire bottoms to the target body's ground bound, a fresh packaged Chrome inspection shows all four wheels in the arches with no page errors. The tire model is still generic; the stock body, window textures, and selection remain intact.
Open or save full-size capture · Inspect the Impreza WRX
The renderer was using the cobble texture alpha as a very faint overlay, leaving most of the roadway visually flat. Materials marked usage 2 / blend 0 now render their recovered neutral texture RGB, with vertex prelighting reduced to luminance for cobble surfaces. The served default view reached “City scene ready” with no browser errors and the road texture is now visible. The exact road width, centerline layout, lighting balance, and building placement still need comparison with retail gameplay.
Open or save full-size capture · Open the default drive view · Compare the previous material composite
Chrome drive trace on route 716: W+A for 1.5 seconds, then W for 2 seconds. The car follows the first bend, then leaves pavement on the next curve and stops beside the storefront. This exposes a route-following and steering limitation in the current prototype; it is not evidence by itself that the storefront transform is wrong. The latest run also softened low-speed keyboard steering. This drive is still unfinished.
Open or save full-size capture · Reopen the tested route start
The exporter bakes each body object's transform into its vertices, but the old manifest described axes from before that bake. Traffic now uses the transformed Z-up axis and the transformed forward sign. In this 1.8-second throttle capture the player reaches 28 mph, stays supported by the road mesh, and the browser reports no errors. This verifies the arterial stretch and visible traffic orientation; it does not establish correct placement on every route.
Open or save full-size capture · Open this arterial view
The stock garage manifest populates 29 selectable retail cars. This packaged city capture selects the Nissan Skyline GT-R R34, shows its original engine recording status, and records a short drive at 20 mph with no browser errors. This verifies a non-default car in the drive loop; each model's wheel fit and engine sample still needs its own review.
Open or save full-size capture · Drive the Skyline · Inspect the stock build
Fresh 1280 × 720 capture from the packaged viewer in headless Chrome. The city reached “City scene ready” with 35,188 nearby triangles and 146 nearby textures; the WebGL context remained available through vehicle upload and the page reported no runtime error. The controls start collapsed so they do not cover the road or car; “Show controls” opens the car, region, and traffic settings. The skyline and distant traffic are visible, while vehicle and street detail still need further work.
Open or save full-size capture · Open the packaged driving view
Fresh 1280 × 720 packaged Chrome capture. The city reached “City scene ready”; the 120-unit neighborhood indexed 6,964 nearby building-face triangles for contact checks with no page errors. Ambient cars are present on the roadway, though their shape and orientation still need a closer per-model review. A separate route 1055 run reached a corridor wall and stopped without losing road support.
Open or save full-size capture · Open the packaged drive
The pickup's bottom-face normals and thin bounds exposed a bad axis assignment: its traffic BODY_A is Y-up, but the viewer had mapped Z upward. The manifest now uses the geometry-confirmed X-long/Y-up/Z-side frame (BUS remains Z-up), so the body sits upright and the authored tire axles map to the body side. Fresh packaged Chrome reached “City scene ready” with no browser errors. Paint and body detail still look rough, and the whole fleet needs per-model visual review.
Open or save full-size capture · Open the packaged drive
Fresh 1280 × 720 packaged Vulkan Chrome view. The L4RA scene reached “City scene ready,” rendered four nearby traffic bodies, and reported no browser errors. The body axes now follow the geometry-confirmed up direction in the normal drive view too. The cars are still small in this frame, and their paint detail, exact routes, and appearance at closer range remain unfinished.
Open or save full-size capture · Open the packaged drive
Each eligible traffic body was loaded alone in packaged Vulkan Chrome, using the same street and a side view. All reached the scene-ready state without page or console errors. The mesh evidence supports X-long/Y-up/Z-side for these models; the contact sheet also makes the remaining coarse surfaces, material gaps, and approximate wheel placement visible. This is a diagnostic sweep, not a visual sign-off.
Open or save full-size contact sheet · Open the packaged drive
Side-by-side packaged Vulkan Chrome captures from the same default view. The broad orange bars came from treating alpha-bearing lane and road-strip textures as opaque. They disappear when those textures blend over asphalt, and the generated dashed divider remains. The surface is now lighter and less cluttered, but most retail lane markings are still missing and the overall road appearance needs calibration. Runtime reached “City scene ready” with four traffic bodies and no page errors.
Open before capture · Open after capture · Open the packaged drive
Lane and intersection textures contain complete asphalt tiles with painted lines in RGB; non-cobble road strips use alpha over the asphalt base. The viewer now applies those paths separately, and route-generated center dashes are omitted where an authored lane marking texture covers the road. Chrome reached “City scene ready” with four traffic bodies and no page errors. The center carriageway still looks broad and the player's position relative to its yellow lines needs a route-by-route check.
Open or save full-size capture · Open the packaged drive
Fresh packaged Chrome view at route 450, distance 300. The crosswalk texture and road surface render, and the approach is open to traffic, but the scene transitions from a broad unmarked intersection patch to narrow marked lanes. This checkpoint confirms the markings are present; it does not confirm that route center and authored lane direction line up through the junction. A separate route 450 distance 100 load lost its WebGL context during backdrop upload, so this route still needs a reliability check.
Open or save full-size capture · Open this route checkpoint
Fresh packaged Chrome capture at the default arterial start, route 450 distance 560. The lane and traffic are visible and the browser reaches “City scene ready” without JavaScript errors. The player car still appears to overlap a yellow boundary line, so the start placement is not validated as drivable-lane aligned. This is a visible remaining defect, not a completion capture.
Open or save full-size capture · Open the default drive
The earlier route 450 distance 560 start placed the 240SX over the source yellow divider. A rendered offset sweep compared 3, 6, and 8 world units; 7 is now the default for this L4RA arterial start. The source yellow divider is left of the car on the gray carriageway, and the fresh packaged scene reached “City scene ready.” This is visibly better lane placement; route-scale and curb clearance still need checking.
Open or save full-size capture · Open the updated default drive
In headless Chrome, holding W moved the 240SX from 0 to 27 mph without a page error. After releasing the key, the follow-up capture shows the car still rolling at 10 mph on the paved approach, outside the yellow-marked center strip. This verifies a short drive at the corrected start; it does not yet verify steering through the intersection or the full route.
Open or save full-size capture · Drive the updated build
The 240SX now starts at route 450, distance 480, with 7 m lateral offset and 140 m of checked pavement ahead. A fresh packaged SwiftShader Chrome run held W for 3.5 seconds; the runtime reached 50 mph and the captured HUD shows the car still accelerating on the asphalt. No road or building collision or page error was recorded. This is a short straight-line check; steering through the junction and driving beyond the route fragment remain unresolved.
Open or save full-size capture · Drive the packaged build
The traffic renderer samples the road mesh beside each route and uses the first right-hand RDP_LANEB/C/D lane tile when one exists. In this L4RA scene, route 587 and 586 cars use a 6.5 m offset onto lane-marked pavement; route 711 and route 450 cars stay at 1.5 m where the nearby surface is an intersection tile. The full 240-unit package reached “City scene ready” with 106,069 nearby triangles, 238 textures, and no page errors. The crosswalk traffic still needs a junction-by-junction placement check.
Open or save full-size capture · Open the full-radius default drive
The stock wheel meshes now turn as a geometry-based Ackermann pair using the recovered axle anchors, and rotate according to distance traveled and exported tire radius. Chrome’s SwiftShader capture reached “City scene ready” with no page errors; the debug driving state confirmed a 0.55 rad stationary steer command, and a separate 1.8-second throttle/steer trace produced nonzero wheel rotation. The city view is fully textured here. This improves the visible car response but does not reproduce the retail tire-force physics; road placement and body finish remain under review.
Open or save full-size capture · Open the packaged city drive
The viewer now samples the region's extracted dusk panorama for paint, glass, and chrome reflections and uses the nine nearby authored streetlights in the vehicle pass. Headless Chrome reached “City scene ready” with no JavaScript or WebGL errors. The side glass and body highlights read more clearly than in capture 131; the wheel faces still look too dark, and the stock vehicle finish needs more work.
Open or save full-size capture · Open the packaged city drive
The 240SX's native wheel chrome slices use alpha blending over the sidewall. The viewer now keeps that material opaque for the wheel face, preserving the exported rim geometry and texture while the tire remains textured. In headless Chrome both wheels show their stock spokes, the front wheels retain their steering angle, the scene reaches “City scene ready,” and the page reports no errors. Other car models still need a close wheel and material review.
Open or save full-size capture · Open the packaged city drive
The keyboard drive update now uses the retail executable's recovered frame subdivision rule, with equal steps no longer than 4.4 ms. A four-second W input moved the car along route 716 in headless Chrome's SwiftShader WebGL2 path with no JavaScript or WebGL errors. This image was captured while only 56 of 148 nearby textures had uploaded; the pale blocks around the overhead flower lights are missing-texture placeholders, not evidence of misplaced building geometry. A new progress indicator stays visible while those textures finish.
Open or save full-size capture · Open the tested route
Fresh packaged SwiftShader Chrome capture at the default L4RA start. All 137 nearby textures finished uploading, the runtime reports 31,078 nearby triangles, and there are no browser errors. The overhead fixtures resolve into the authored flower-light decorations. The broad intersection surface and blurred road detail remain visible limitations; this is a fully textured review frame, not a completion claim.
Open or save full-size capture · Open the default city drive
The viewer now shows a top-right upload counter after the first driveable scene appears. This capture records 60 of 137 nearby images uploaded; the loading card has closed, the car is drivable, and the remaining pale areas are unfinished textures. The counter disappears when all required nearby images arrive; capture 135 shows the completed default view.
Open or save full-size capture · Open the packaged city drive
The city is already playable while normal-quota traffic assets finish uploading. The corner indicator now shows the traffic model stage after the nearby city textures complete; this packaged Chrome capture records 2 of 12 traffic bodies loaded with no browser errors. It clears automatically when the roster is ready.
Open or save full-size capture · Open the city drive
The packaged city page exposes all 29 stock cars. Mustang GT, Skyline, and Escalade were selected and loaded while the nearby city texture stream continued. This standalone Escalade capture reached “stock model ready” with 103 selected part meshes, 5,117 triangles, three embedded textures, and no browser errors. A city-page deep link selected the same Escalade, enabled the car selector before traffic finished loading, and moved at 3.5 m/s under W with its original engine recording active and no page errors. The screenshot verifies the stock asset and glass material in inspection mode; the city view is shown in item 140.
Open or save full-size capture · Open the city drive
Road-triangle samples showed the previous camera sat over the transition from the intersection tile to `RDP_LANEB`. The default route 450 start moved 15 m farther along the lane to distance 525, with 90 m of supported pavement remaining in that route fragment. This Chrome capture shows continuous gray pavement around the car, the yellow center line and white dashed divider, plus ambient traffic. A 2.5-second W input at this spawn moved the car 3.1 m at 6.6 m/s with no reported road/building collision or browser error. Other streets and the broader city layout still need visual review.
Open or save full-size capture · Open the updated default drive
The traffic axis transforms have negative handedness for most recovered car models. Their triangle winding is now corrected before upload so WebGL back-face culling keeps the body shells visible. A fresh packaged capture reached the ambient vehicle roster with the full city texture stream complete; several cars are now visibly shaped and upright on the road. Their spacing, wheel contact, detailed orientation, and the bright near-road lighting still need review.
Open or save full-size capture · Open the packaged city drive
Fresh packaged Chrome capture after all 245 required road and city texture slots completed (one unused slot is skipped) and the ambient vehicle roster loaded. The yellow center line and white lane dash appear in the authored lane positions, and traffic vehicles now render on the road. This frame still shows a broad bright near-road area, sparse traffic, and traffic body shapes that need a closer orientation and material audit; it is evidence of current state, not a completion image.
Open or save full-size capture · Open the packaged city drive
Road-material sampling found the old start was still on a broad intersection tile and the 7 m offset straddled the `RDP_LANEA`/`RDP_LANEB` boundary. The default now starts at route 450 distance 510 with an 8.5 m lateral offset, centered in `RDP_LANEB`; the white dashed lane edge is visibly left of the car. After the full scene reached “City scene ready,” a three-second W input moved the car 3.5 m on the pavement with no road, building, traffic, or browser error. The broad yellow center strip and low-detail road texture remain unresolved.
Open or save full-size capture · Open the updated default drive
A headless Chrome run on L4RA route 134 reached 130 mph in sixth gear with road support and no browser errors. This capture is from the coast immediately after that check, while nearby textures were still streaming; the HUD shows 35 mph. The road is visible again, but the lane placement and scene coverage through this stretch remain under review. This is not a completion image.
Open or save full-size capture · Open the tested straight
This live packaged Chrome capture reads 130 mph, sixth gear, and 6,760 RPM on L4RA route 134. The road and nearby buildings are visible during streaming, the road sampler reports support, and the page has no errors. The 240SX appears too close to the yellow center divider, and this long-straight diagnostic route does not establish clean citywide road continuity.
Open or save full-size capture · Open this tested route
Packaged Chrome capture of L4RB after its source road texture set uploaded. The 240SX and street are visible while nearby city textures continue streaming; the captured loader overlay is in its fade-out frame. This confirms the viewer no longer waits for every nearby texture before revealing the drive. No ambient traffic appears in this frame, so it does not visually verify the corrected traffic model axes; that check remains open.
Open or save full-size capture · Open this streamed scene
Fresh SwiftShader WebGL2 capture at L4RA route 450, distance 300. The crosswalk and lane paint are from the recovered road textures; the viewer no longer draws a separate road-marking geometry pass. The traffic loader selected COUPE and PARCELVAN on this segment and rendered three nearby bodies. Two still look misoriented or backward from this view, so the traffic orientation issue remains open and needs a closer model-by-model source check. The scene reached its ready state without a page error; unrelated city textures were still finishing.
Open or save full-size capture · Open this route
Fresh SwiftShader WebGL2 capture on L4RA route 450. The extracted route direction vectors oppose increasing waypoint order on all 1,154 routes examined, so the viewer now samples traffic in the decoded direction and retains source waypoint IDs for junction links. In this frame the blue, yellow, purple, and red traffic bodies sit on their wheels on the authored intersection pavement. The crosswalk and lane paint come from the recovered textures; there is no separate generated road-marking pass. This verifies the tested junction, while the full fleet and all traffic transitions still need review.
Open or save full-size capture · Open this route
Fresh capture from the packaged viewer with four extracted traffic models enlarged for inspection. Their body bottoms and wheel bases meet the road in this sample, and the crosswalk remains part of the recovered road texture. The source geometry audit reports the bottom-face normals aligned with each model's mapped up axis. This does not prove the entire fleet or every route orientation is fixed: the extracted traffic bodies still look like simplified, flat previews with crude wheel placement. A lighting pass now applies nearby street lamps and player headlights to traffic; the close-up shows that this alone does not resolve the model fidelity issue.
Open or save full-size capture · Open the tested intersection
Unedited 1280 × 720 capture from the packaged build at the default 240SX start. The visible lane and center markings are part of the recovered road textures, and ambient cars appear ahead. The viewer reports no browser errors, but this frame was captured while two city textures were still uploading; the lighting and distant traffic remain visibly weak. It is a current-state check, not completion evidence.
Open or save full-size capture · Open the packaged drive
Fresh packaged Chrome 150 capture with BODY_B forced for PARCELVAN. Its source frame reverses forward and up signs relative to BODY_A; the viewer used BODY_B's measured axes and rendered the van upright on the road. The crosswalk is part of the recovered road texture. The BODY_B shell is visibly low detail, and this diagnostic does not establish the retail four-slot LOD mapping or fix the overall traffic presentation.
Open or save full-size capture · Open this diagnostic view
Fresh 1280 × 720 Chrome capture from the current source viewer. The runtime reports zero generated road markings; the lane paint and crosswalk use the city’s recovered RDP materials. Four nearby cars are upright in this view and their model bases match sampled road height within 0.001 world units. Route positions already on a recovered lane surface are no longer shifted by a fixed 1.5 m. Their bodies remain visibly flat, their wheel placement looks crude, and this does not verify every traffic model or route. Texture streaming was still at 92 of 96 when captured; there were no browser errors.
Open or save full-size capture · Open this route
Composite of twelve individual WebGL captures, one for each model with a positive normal-density quota. The camera was recentered on each moving route pose; all twelve BODY_A bodies appear right-side up in this view. Several tires still render as plain black faces, and COUPE/SUV tire proportions need closer comparison with their body arches. This checks model orientation and mesh visibility only; it does not prove traffic drives in the correct lane or matches the retail part assembly.
Open or save the full fleet sheet · Open the packaged drive
Fresh 1280 × 720 SwiftShader WebGL2 capture after updating the viewer and asset cache keys. The viewer reports zero generated markings; this frame's crosswalk and road paint come from recovered city materials. Four traffic bodies on routes 790/791 sit at the sampled road height and appear upright in this intersection. This does not clear other routes or fix the visibly crude traffic body and tire assembly. Browser errors: none.
Open or save full-size capture · Open this route
Decoded directly from the installed L4RA stream TPK. The first two panels show the RDP_LANEA yellow edge stripe and RDP_LANEB white edge stripe after the same P8 channel handling used by the viewer; the third shows the parking-line map. No lane markings are synthesized in the viewer. This checks the source pixels and channel mapping, but does not yet verify every tile's UV placement against retail gameplay.
Open or save the texture comparison · Open the COUPE-only traffic diagnostic
Captured from the source viewer after the base city texture set reached 96/96, then holding W for 4.5 seconds. The HUD reached 38 mph. Chrome's live debug data reports four COUPE bodies rendered on route 450 with their bases 0.0043 world units above sampled road surfaces and route-direction dot values near -0.988. The bodies are upright, but the estimated TIRE_FRONT_A disks extend above their roofs and look like black slabs. This pins the visible fault on tire size or mounting; the retail tire choice and attachment data remain unconfirmed. Browser page exceptions: none.
Open or save full-size capture · Open this COUPE-only diagnostic
Both panels use the recovered route 450 scene, the same COUPE BODY_A geometry, and all 96 city textures. The A panel places black disks above the roofs; the B panel uses the smaller authored TIRE_FRONT_B meshes and keeps them below the body line. The WebGL preview now defaults to B and accepts trafficTireVariant=A for comparison. This improves the visible assembly with estimated mounts; the retail loader's exact A/B slot choice and wheel attachment transforms remain open. The road's yellow/white dashes and crosswalk are sampled from RDP textures; the renderer generates no marking geometry.
Fresh Chrome capture from the source viewer at route 450, distance 300. It shows four normal-quota traffic cars and separate source triangle sets for intersection, yellow lane, neutral lane, and shoulder materials, with source UVs preserved. Follow-up reconstruction of the intersection material slice shows the broad crossing bars come from the P8-swizzled RDP_INTERSECT_AA_WH texture mapped by the original UVs. The viewer generates no marking geometry. This frame still does not verify every road tile or traffic route.
Open or save full-size capture · Open this route
Fresh Chrome captures from the packaged build, one for each positive-quota traffic model, using BODY_A and the current TIRE_FRONT_B preview. All twelve bodies appear upright in this camera view, and the browser reported no errors. Retail executable data identifies cullable tire corners in FL, FR, RR, RL order, but those records do not provide the actual wheel attachment transforms. This check does not establish correct traffic direction on every route or direct retail fidelity.
Open or save the full fleet sheet · Open the packaged city drive
The r89 capture uses texture RGB for opaque lane and intersection materials. A later per-slice reconstruction confirmed the crossing bars are in RDP_INTERSECT_AA_WH when sampled with the original UVs and palette channel order. Revision r90 removes the remaining display gain from opaque road RGB. This capture is incomplete and does not verify every city tile or overall visual fidelity.
Open or save full-size capture · Open this route
Contact sheet of the 24 RDP road images embedded in the cleaned city export. The lane edge maps contain yellow and white stripes, and the intersection map's crossing-bar pattern becomes clear when projected through the source mesh UVs. The viewer now leaves opaque map RGB unmodified.
Open or save the texture sheet · Open the route capture
Each panel isolates one material slice from the original TRN_CS_ROADA_CHOP_M1_R3 intersection geometry, using its exported triangle positions and UVs. The upper right RDP_INTERSECT_AA_WH slice reproduces the crossing bars after the source P8 channel swap; the lane slices separately show their yellow and white edge paint. This confirms the marks are texture-mapped source data, not generated overlay geometry. The full route capture still needs comparison with the XP game.
Open or save the material-layer audit · Open the route capture
The decoded L4RA traffic basis opposes increasing waypoint order on 1,153 of 1,154 routes by more than the old confidence cutoff. TrackRoutesA32 (key 852) was the exception: its length-weighted agreement was -0.125, but two of its three segments point against file order. Revision r91 now uses the sign of the measured agreement and reverses this spline to follow the retail basis. This corrects one route-direction edge case; it is not a full traffic-route validation.
Open or save the direction diagram · Open route 852
Fresh SwiftShader Chrome capture from the source viewer at route 450, taken after all 96 nearby city textures uploaded. The decompiled RDP materials have usage=2 and blend=0, so revision r92 no longer interprets their alpha as coverage over an estimated asphalt base. The crossing bars and lane stripes remain source-texture marks; the viewer reports zero generated markings. Traffic was disabled for this road-material check. Road scale, placement across other tiles, and appearance against XP gameplay still need work.
Open or save the full-size capture · Open this route
Fresh 1280 × 720 SwiftShader capture after all 96 nearby textures uploaded, with the source TAXI model forced for diagnosis. Revision r93 uses the decompiled waypoint forward vectors to orient traffic on routes 790/791. Runtime diagnostics show four taxi poses following the decoded direction basis; their model bases are about 1.5 cm below sampled road height. This checks one model and one junction only. The full traffic roster, other routes, exact retail vehicle assembly, and XP visual comparison remain open.
Open or save the full-size capture · Open the TAXI-only diagnostic
Fresh fully textured Chrome view with the camera following a moving COUPE on recovered route 791. The car's body is upright and its rear faces the camera while it follows the decoded forward vector across the intersection; a second COUPE moves farther ahead. This verifies the sampled route and body frame in this stretch. It does not verify the other traffic models, exact retail wheel mounts, or traffic behavior throughout Bayview.
Open or save the full-size capture · Open the COUPE-only diagnostic
The traffic preview now reads the four axle positions from each matching GLOBALB.BUN car record and transforms them through the selected body's authored frame. This capture shows the COUPE on the decoded pavement with those anchors active. It is a review frame, not a completion claim: the low-poly tires still render as dark slabs, the paint is a flat preview color, and the full traffic roster and routes still need inspection.
Open or save the full-size capture · Open the COUPE-only diagnostic
Revision r96 maps each tire's baked source frame into its matching body frame, then uses the four axle anchors decoded from GLOBALB.BUN. The pictured COUPE bodies are upright and sit on the recovered intersection surface. The source tire meshes are still visibly low-poly dark slabs, the body colors remain preview paint, and this capture does not establish that every model and route matches retail gameplay.
Open or save the full-size capture · Open the COUPE-only diagnostic
This capture uses the COUPE source body and tire meshes, GLOBALB axle positions and body/tire transforms, plus the shared CARBOTTOM, DRIVER, WINDOW_TRAFFIC, and TIREBACK maps decoded from GLOBALB.BUN. The windows read as tinted glass and the tires appear round again. Paint remains a fallback preview color because the traffic body-paint hash has no matching image in the shared pack; the full roster still needs retail comparison.
Open or save the full-size capture · Open the COUPE-only diagnostic
The traffic GLBs now retain BGRA prelight from the source 36-byte car vertices as glTF COLOR_0, alongside the shared GLOBALB glass and tire maps and exact body/tire transforms. In this Chrome route capture, the body panels have more source shading while the bodies remain upright. The paint base is still a fallback color, and this is not a retail-game comparison.
Open or save the full-size capture · Open the COUPE-only diagnostic
The same exporter change was applied to the complete stock player-car roster. This is the selectable 3000GT in the city drive with source vertex colors retained. The image documents the current preview; compare it with the XP game's night lighting before treating the material as final.
Open or save the full-size capture · Open the 3000GT drive
This Chrome frame was captured after revision r100 changed lane, intersection, parking, and grill materials to use sampled source RGB directly, without shader-added line geometry, alpha coverage, or estimated asphalt blending. The yellow center marking and white lane markings are texture pixels. At capture time the city texture stream was only 52/221, so the malformed building appearance and distant traffic in this frame are not a completed scene review; traffic orientation across the full route roster remains open.
Open or save the full-size capture · Open the TAXI-only drive
The r101 audit rendered each positive-quota BODY_A at one fixed, recentered pose on route 791. All 12 looked upright in those captures, but that static check missed the upside-down cars now reported during live driving. It does not validate heading changes, every route, or the retail traffic mover’s runtime transform. The source tire meshes also remain low-detail.
Open the scrollable 12-model audit · Open or save this COUPE frame
Current source viewer capture at route 450, without debug markers. There is no generated lane-marking geometry: the visible crosswalk, yellow hatch, and lane paint are drawn from the exported road material textures. The capture reached 92/96 nearby textures and shows three upright traffic bodies at this view, but their distant detail cannot prove the full fleet orientation. Their tire fit and retail placement still need work; road marking placement still needs direct comparison with the XP game.
Open or save the full-size capture · Inspect the isolated authored road material layers
The route-450 frame selected PANELVAN, SUV, HATCHBACK, and PANELVAN; each has a positive source quota in the recovered density columns. It has no browser errors and the car bases remain on the sampled road. This does not demonstrate an orientation fix: these cars are too distant here to settle tire fit or the reported fleet inversion. Only 92/96 nearby textures had uploaded.
Open or save the full-size capture · Open the close 12-body orientation audit
The retail material records mark the road-strip textures as opaque, but the viewer had replayed some of those same slices through a guessed alpha-overlay pass. The normal renderer now draws each source road slice once; the legacy composite remains available for comparison. This fresh route-450 capture reached “City scene ready” with no page errors and reports zero generated markings. The yellow hatch is still visibly positioned across this intersection, so the XP placement comparison and the reported traffic inversion are not resolved.
Open or save the current source-texture capture · Compare the legacy composite diagnostic · Open the tested drive · Read the source and decompilation notes
I focused the live camera on one moving PANELVAN from routes 790/791. Three nearby source traffic bodies are upright in this sample, with bases within 0.001 world units of the sampled pavement and no browser errors. This does not verify other routes or disprove the reported inversions elsewhere. The bodies still look too flat, tire fit remains crude, and source paint state/LOD assembly is unfinished.
Open or save the close traffic capture · Read the verified type-2 render path notes
Revision r106 corrected the axis metadata but left the traffic vertex export in the retail frame because its exporter call still used gltf_vehicle_axes=false. This distant capture is retained as an intermediate diagnostic, not proof that the correction worked. Revision r107 fixes the geometry export frame too; see the following close COUPE capture.
Open or save the capture · Read the axis and retail resource notes · Open this drive
Fresh packaged route-450 capture after revision r107 enabled the final (Y, Z, -X) conversion for all traffic body and tire exports. The corrected axis metadata now matches the actual GLB vertices. Two COUPEs are upright and follow the recovered route in this view; the close car's base is within 0.005 world units of sampled pavement. No browser errors occurred. This checks one model and one intersection. The bodies and wheel fit remain crude, and the complete fleet/routes still need review. The road crosswalk and lane paint use recovered source textures; generated markings remain zero.
Open or save the capture · Read the exporter and decompilation notes · Open this drive
With the exporter now applying the same (Y, Z, -X) conversion recorded by the model metadata, I forced each of the 12 positive-quota traffic models through the live route-450 scene at a close side view. All 12 appear upright, with tires below the body line; no browser errors were reported. This is a one-intersection fleet check and does not validate every route, the retail LOD choice, or paint state. The crosswalk and lane paint remain source road textures, with no generated marking geometry.
Open the scrollable 12-model capture set · Open or save the COUPE capture · Read the retail resource notes
This installed region contains pavement triangles but no RoutesFreeRoam waypoints. The viewer now starts the selected car on a long source pavement triangle and derives its heading from the triangle edge. The package reaches the ready state without a browser error. L4RH remains a sparse scene in the recovered export; this start does not fill in missing scenery or create traffic routes.
Open or save the capture · See the L4RR track start · Read the current region notes
In the packaged build, the 240SX moved about 29 world units during a 3.5-second W input on route 450. The player stayed on sampled pavement, with no road-failure diagnostic, collision diagnostic, or browser error. The road stripes in frame come from source materials; ambient cars use the corrected glTF axis export. This checks one short stretch, not every road or the XP gameplay match.
Open or save the drive capture · Open this route start · Read current route and material notes
This fresh normal-mode Chrome capture uses the current packaged viewer after changing the script and asset cache keys past r103. Its source-route traffic sample shows three bodies upright on their wheels at this intersection; a separate runtime readout selected four positive-quota cars on RoutesFreeRoam routes 790/791. The current renderer reports zero generated road markings. The yellow hatch, crosswalk and lane paint come through the authored road material textures/UVs. Paint matching, detailed tire fit and orientation across the full route set are still unresolved, so this is a checkpoint rather than a completion claim.
Open or save the full-size capture · Inspect the source road material layers · Read the decompilation and source notes · Open the normal-mode scene
This normal-mode Chrome capture uses the packaged r113 viewer. Traffic is no longer moved sideways by searching for nearby lane-texture material names; each body follows the decoded RoutesFreeRoam waypoint position and heading. The hatch marks, crosswalk, and lane paint remain embedded source road textures. This screenshot checks one Bayview view only: the retail FuturePoint lane-phase selector, full route direction behavior, and visual match to XP gameplay are not yet reproduced.
Open or save the capture · Read the lane-action decompilation notes · Open this route start
This software-WebGL capture reached “City scene ready” with all nearby textures uploaded and no browser errors. The default path reports no generated markings; the visible road paint comes from the embedded source road material textures. The stripe positions still need a retail gameplay comparison, and no nearby traffic is visible in this camera frame, so it does not settle the reported fleet inversions.
Open or save the full-size capture · Read the current source and decompilation notes
These are the decoded retail L4RA road material images embedded in the exported city GLB, extracted from the XP-installed game files. The viewer reports zero generated road markings and samples these textures directly. This verifies the paint source; it does not yet verify that the road meshes use the correct material slice, UV orientation, and placement for every street.
RDP_LANEA source texture · RDP_INTERSECT source texture · RDP_PARKING source texture · Current traffic scene capture
This packaged-page capture was taken from the server on port 8095 at 8/96 nearby textures, so it is not a finished gameplay frame. Even with that limitation, the nearby building geometry is visibly malformed and the 240SX still looks too flat. I rejected the rebuilt skyline export after its preview rendered black. The source road textures are confirmed; road placement, building assembly, and moving-fleet orientation remain open.
Open or save the full-size capture · Open the current drive · Read the matching-decompilation status
I forced each of the 12 positive-quota source BODY_A models through the live moving traffic renderer on L4RA route 450. They follow recovered waypoints 790/791; the runtime readout kept their bases within 0.001 world units of sampled road height. All 12 appear upright in this sample. That narrows the problem but does not clear other routes, junction transitions, or the reported user view.
Open the scrollable 12-model live audit · Open or save the COUPE frame · Read the lane-action decompilation notes
This screenshot waits for the page's ready flag instead of capturing the loading frame. The scene reports 90,091 nearby triangles, 832 routes, and 12 traffic models loaded; four traffic cars rendered in this frame. The lane paint comes from the source road textures. This view does not verify every street or reproduce the reported upside-down cars, so the export is still incomplete.
Open or save the ready-state capture · Read the source-bounds rotation audit · Review the limited fleet orientation sample
This is a fresh Chrome capture from port 8095. The visible lane stripes and yellow center paint are sampled from the retail road textures; the renderer reports no synthetic marking geometry. Their placement still does not match reliably across the exported street meshes. The same view confirms the building assembly is visibly malformed, and only 3 of 12 traffic models had loaded when the screenshot was taken, so it cannot verify traffic orientation. I am keeping these defects open rather than treating the earlier route sample as completion.
Open or save the full-size capture · Compare the decoded source road layers · Review the limited 12-model traffic sample
Fresh 1280 × 720 Chrome capture from the packaged page after fixing the split-city manifest path and creating WebGL after the large city file is indexed. The page reached “City scene ready” with 96/96 nearby textures, 12 ambient models, four traffic cars in frame, and no browser errors. The visible road paint comes from source RDP textures. This checks one route and intersection; citywide building placement, road-mark alignment, and the reported upside-down traffic remain open.
Open or save the full-size capture · Open this route · Read the traffic decompilation notes
I forced all twelve positive-quota BODY_A models through the moving traffic renderer and centered the camera on the active cars. Their body shells stay upright in this route 450 sample; the black tire disks remain visually crude and can make vehicles read as inverted at distance. This does not verify every route transition or reproduce the reported view. The road hatch, crossing bars, and edge paint are sampled from the source RDP textures; their placement across every exported street still needs comparison against the XP game.
Open or save the full-size 12-model audit · Inspect the source road texture placements · Read the traffic FuturePoint decompilation notes
Fresh packaged Chrome capture from the live WebGL renderer, viewed from above to inspect the road materials. The double yellow center stripe and white lane dashes are from the source RDP texture slices and UVs; no line geometry is generated. This confirms the renderer is sampling original paint, but it does not establish that every road slice is attached to the correct street or matches the XP game. This road-focused capture does not establish traffic orientation; route-wide backward or inverted traffic remains unresolved.
Open or save the full-size capture · Compare source road material layers · Read the retail traffic decompilation notes
The live route-1016 car advances onto route 1203 at waypoint 27, then travels toward source waypoint 26. Its sampled heading is about −0.12 radians, matching the decoded route basis and waypoint positions. This confirms the weakest L4RA geometric join is a sharp but forward source connection in this sample; it does not explain the reported inversion or validate other junctions. The visible traffic bodies are upright here, while tire/body fit and city scene defects remain obvious.
Open or save the full-size capture · Read the transition audit
The live intersection street mesh is TRN_CS_ROADA_CHOP_F4_R1, with source RDP_INTERSECT, RDP_LANEA, and RDP_LANEB texture slices. The crosswalk and lane paint come from those textures and source UVs; the viewer generates no marking geometry. Their placement still needs a retail-game comparison. Traffic telemetry showed four cars within 0.052 world units of the sampled road surface; their shells were upright in this capture, which does not reproduce the reported inversion elsewhere.
Open or save the full-size capture · Inspect the source road material layers · Review source placement/prototype audit · Read the source-object audit notes
Captured from the packaged page on port 8095 after the full city scene loaded. The yellow edge stripe and white dashes are sampled from retail RDP material images and UVs; the runtime reports zero generated marking geometry. The lane placement still needs a same-location XP gameplay comparison. The visible moving traffic is aligned with its recovered route basis and appears upright in this sample; that does not resolve the reported route-wide backward/upside-down cars.
Open or save the full-size capture · Inspect the decoded source textures · Read the route-basis and FuturePoint decompilation · Open this start in the viewer
Route 450 reached “City scene ready” with the user's trafficBodyVariant=B query present but diagnostics disabled. Network capture confirms the renderer loaded only the high-detail source BODY_A assets. The decompiled renderer selects detail slots by projected size, but the slot-to-BODY_A/B/C mapping is still unknown; normal play no longer forces an unverified variant. This prevents that query from inverting affected body variants, but route-wide traffic direction and road-paint placement remain open.
I forced 4DR_SEDAN02 BODY_B through the moving traffic renderer and centered the camera on its route-744 pose. The shell is upright and the base is within 0.002 world units of the sampled road in this frame. This rules out a universal BODY_B axis inversion, but it does not reproduce the reported cars or clear BODY_C and route transitions. Cyan crosses are debug pose markers.
Open or save the full-size capture · Read the source body-axis notes · Open the B-variant diagnostic scene
I centered Chrome on four live cars on routes 744, 741, 739, and 737. Their BODY_A shells are upright in this sample. The current GLB assigns the intersection, lane A, and lane B source textures to this road mesh; the viewer reports zero generated markings and samples the opaque texture RGB directly. This does not show that each road material slice is attached to the correct tile or reproduce the reported inverted traffic elsewhere.
Open or save the full-size capture · Read the r120 material and decompilation notes
This fresh r122 capture is a live L4RA intersection view after removing the guessed BODY_A/B/C distance switch. Telemetry reports four traffic cars on routes 790/791, all using BODY_A; the scene reached “City scene ready,” reported zero generated markings, and had no page errors. The crosswalk, hatch, and lane paint still come from the RDP textures, but this frame makes clear their placement and the overall roadway presentation remain wrong. This does not clear inverted cars on other routes or reproduce the XP view.
Open or save the full-size r122 capture · Read the r122 decompilation note · Open this current drive view
The route 450 camera covers source mesh TRN_CS_ROADA_CHOP_M1_R3, with its crosswalk, lane paint, and hatch mapped from RDP_* texture slices. The older r122 yellow hatch was not a generated marking; it came from a red/blue channel swap applied by WebGL to the retail P8 palette. Revision r123 removes that conversion. Tile placement and the overall road layout still need work.
This capture is from a separate clean-room renderer reading the same installed game data at route 450's world location. Its blue/cyan hatch agrees with the raw RGBA palette values in RDP_ROADSTRIP_MC. The previous WebGL shader swapped red and blue for P8 textures and made those pixels yellow; r123 now samples palette RGB directly. This is a data-renderer comparison, not an XP gameplay screenshot, and it does not settle road-tile placement or the upside-down traffic reports.
A fresh 1280×720 WebGL capture now shows the blue/cyan hatch from the retail P8 palette and four moving BODY_A cars on routes 790/791. The car bases are within 0.015 m of the sampled pavement, and these four shells are upright in this pose. The first traffic refresh now runs immediately, even when the browser clock is below the 1.8-second refresh interval. This only checks these two local routes; road assembly and route-wide inversion remain open.
Open or save the full-size capture · Read road texture and placement evidence · Read the original traffic decompilation notes
This is the 64×512 PNG embedded in city-l4ra-textured-clean.glb, traced back to the RDP_ROADSTRIP_MC TPK record in the extracted STREAML4RA.BUN. Its cyan hatch is present in the source image itself; the viewer samples it through the road mesh UVs. This confirms the marking pixels are texture content. The mesh/UV placement on the street can still be wrong.
I moved the debug camera onto the four live vehicles in the current WebGL build. Runtime telemetry identifies PANELVAN, SUV, HATCHBACK, and PANELVAN, all BODY_A, on routes 790/791; their body bases are within 0.001 world units of the sampled pavement. The shells look upright in this local view. This narrows the issue but does not explain the user's upside-down or backward cars on other routes.
Open or save the full-size traffic capture · Read the route-frame and retail-resource notes
Revision r127 applies the retail MODULATE2X-style vertex prelight to opaque road textures using the exported COLOR_0 values. In the fresh 1280×720 capture, the pavement is visibly darker and closer in tone to the same-location clean-room retail-data reference, while the blue/cyan hatch, crosswalk, and white lane paint still come from the decoded source textures and UVs. The viewer generates zero marking geometry. This does not correct the placement/layout of the intersection pieces or verify traffic orientation across the city.
Open or save the full-size capture · Compare with the retail-data reference · Read the road texture, UV, and prelight evidence
The extracted route waypoint placed a car on RDP_ROADSTRIP_MC, the cyan hatched divider. Sampling the same road cross-section found RDP_LANEA 2–5 m to the car's right. The viewer now centers traffic on that source lane span when the route origin is on this divider. The cyan marks remain from the original texture, and the car base is within 0.001 world units of the sampled pavement. This checks one route and does not reproduce the reported upside-down cars or confirm the full layout against XP gameplay.
The WebGL and clean-room reference captures focus the same L4RA intersection from the XP-extracted stream. The plaza signs, storefronts, trees, signals, crosswalk, and road pieces appear in broadly matching positions. The WebGL pavement remains brighter and smoother, and its ambient cars still look too plain. The reference is OpenUG2 using extracted retail data, not an XP gameplay screenshot; this checks one block only.
Open the WebGL capture · Open the extracted-data reference · Read the camera and source details
Packaged Chrome capture after all 96 city textures uploaded. Revision r131 removes the custom vertex-color tint from opaque source RDP materials: their decoded RGB and exported UVs supply the paint, with no generated lane lines. The cyan hatch is visibly present in the extracted RDP_ROADSTRIP_MC texture itself, so this confirms pixel provenance but does not prove the original city places the strip here. The four nearby cars look upright in this L4RA sample; the reported inversions on other routes remain unresolved.
Open or save the full-size capture · Read the source-data and traffic audit
Chrome capture of the L4RA route 852 ramp. Runtime data places the visible traffic on decoded route 853; the two cars face along the ramp and look upright in this frame. Other nearby routes are at lower elevations and outside this camera view. This still does not explain the reported inversions elsewhere in Bayview.
Open or save the full-size capture · Read the route and body-frame evidence
This fresh packaged Chrome capture shows the source white lane dashes and cyan median hatch in the original road textures; the viewer generates no marking geometry. The retail route waypoint itself runs down the divider, so traffic is shifted to the adjacent authored RDP_LANEA triangles using a source-geometry/material check. The cars in this view are upright, but the sideways placement still approximates the retail FuturePoint candidate path and does not settle reports from other routes.
I forced each quota-eligible retail BODY_A model through the same route transform and road pose. The 12 captured bodies stayed upright, with each base within 5 cm of the sampled pavement in this route sample. This rules out a consistent BODY_A axis flip for these models here; it does not cover every route transition, the retail LOD binding, or the user's other reported locations. Scroll through the 12 model captures · Read the source frame and decompilation notes
Fresh packaged Chrome capture after all 112 city textures uploaded. Route 106's source waypoint lies on the median, so the player and traffic now use the adjacent exported RDP_LANEA pavement. The player correction is 4.25 m in this sample; the white dashes and cyan hatch are texture pixels, with no generated road paint. This fixes this route's initial position, while exact retail FuturePoint weights and upside-down traffic reports elsewhere remain open.
Revision r135 feeds the source waypoint forward and lateral vectors into the traffic transform, so cars can pitch with ramps. Live telemetry showed seven BODY_A cars on routes 736–744, no browser errors, and body bases within 0.06 m of sampled pavement. The road renderer reports zero generated markings; the blue center stripe and white dashes are sampled road texture pixels. This level-road capture does not resolve the user's citywide upside-down reports or prove that these source road slices occupy the same positions as in XP gameplay.
Open or save the full-size capture · Read the r135 source frame and lane-controller notes
The prior capture showed the source RDP_LANEA_YS marking in cyan. Its material record identifies a P8 palette, whose little-endian Direct3D D3DCOLOR bytes are BGRA; treating them as RGBA reversed red and blue. Revision r136 swaps channels only for recovered P8 materials, restoring the texture-authored yellow stripe while leaving DXT road maps unchanged. The screenshot is a fresh packaged Chrome render with all 114 nearby textures uploaded and no browser errors. The separate L4RD freeway view shows the same correction on the source-textured yellow hatched median. These captures confirm the palette fix, not complete road-slice placement or citywide traffic orientation.
Fresh capture of the current viewer page. The yellow median hatch and white lane dashes are sampled from the embedded RDP lane materials and their mesh UVs; the viewer draws no lane-mark geometry. Their placement and the traffic/body assembly still need retail comparison.
Open or save full-size capture
Packaged Chrome capture after the traffic transform now projects the interpolated lateral vector perpendicular to the forward vector and rebuilds up. Six cars render on routes 736–745; telemetry places five body bases within 0.06 m of sampled pavement. The bodies remain untextured white with crude black tire disks, and this one intersection view does not establish citywide direction or orientation.
Open or save full-size capture · Read the r137 frame notes
Revision r138 fixes a contradictory view filter: cars spawned 10 route units ahead were rejected by a 12-unit minimum. The cutoff now matches the 8-unit in-view corridor. Live route telemetry shows four cars following the recovered forward direction on route 106, with body bases within 0.05 m of sampled road. This capture was taken while the remaining traffic models were still loading; it does not resolve paint, tire quality, or citywide traffic orientation.
Open or save full-size capture · Read the route-106 diagnostics
The source BODY_A up signs agree with all 15 wheel-anchor sets, and the cars in this route 106 frame are upright. The visible result is still wrong: several cars stack into one narrow stream, and the closest body reads as a box rather than a finished retail car. Routes 106 and 116 share waypoint coordinates but have different decompiled selector banks. The current material-based correction puts both streams in nearly the same lane; runtime FuturePoint weights and retail part assembly are still unresolved. Road markings come from RDP textures; no line geometry is generated.
Open or save full-size capture · Read the axis and route evidence
This fresh packaged Chrome capture shows the actual L4RD route 106 scene. Diagnostics report zero generated marking segments; the yellow hatch and white lane lines are sampled from the road materials and UVs. The Coupe is upright in this sample, which does not explain the inversion reported on other routes. Its untextured body and crude wheel openings remain visibly wrong, and the lane shift still uses a temporary pavement-material approximation while the decompiled FuturePoint weights are unresolved.
Open or save full-size capture · Read the FuturePoint selector decompilation
The decompiled waypoint width and signed phase-adjustment bytes now supply the median offset using the original 0x005dd490 scale. The player starts on adjacent pavement in this route sample. Yellow hatches and white dashes still come from the embedded RDP textures and exported UVs; the viewer generates zero marking geometry. The runtime selector and smoothing are not fully recovered, traffic still needs lane-by-lane review, and the reported upside-down cars remain unresolved.
The retail trace says logical phases are 2–5; only one FuturePoint subtype remaps them to bytes 6–9. Revision r147 stops searching both banks without knowing the live subtype. The source RDP textures supply the visible yellow bars and white dashes, and the renderer reports zero generated marking geometry. In this screenshot the yellow bars still look misplaced relative to the drivable lanes, and the barriers/buildings make the scene visibly unlike gameplay. Three sampled traffic cars are upright here; this does not clear the citywide upside-down report.
Open or save full-size capture · Read the FuturePoint selector trace
The route-101 start previously straddled the double-yellow marking. The viewer now samples the original lane texture around the route and applies a decompiled positive corridor candidate only when the candidate clears a nearby paint pixel and lands on authored lane triangles. This capture shows the player in the right-hand lane; the paint itself still comes from the source texture. It is one route sample and still approximates the live FuturePoint selector. Traffic orientation remains under citywide review.
Open or save full-size capture · Read the decompilation and placement notes
The viewer now restricts its direct FuturePoint lane hints to selectors whose route header availability byte is nonzero, as checked by retail helper 0x0040f4a0. The source RDP materials provide all visible paint; diagnostics report zero generated markings. This still does not reconstruct the final traffic lane position because the runtime candidate values and FuturePoint weights are absent from the static route export. Cars in this L4RA sample are upright; the reported inversions elsewhere remain unresolved.
Fresh Chrome close view of a COUPE BODY_A on route 450. Its authored window, underbody, driver, and tire maps are applied from the embedded GLB textures; the body paint remains a flat source fallback because the per-car paint hash is absent from GLOBALB.BUN and local TEXTURES.BIN is empty. The body is upright in this sample. A separate route-order audit supports rejecting reverse-direction endpoint joins, but the viewer still animates route samples at a guessed constant speed and does not run the retail traffic controller.
Open or save full-size capture · Read the texture and controller evidence
Fresh Chrome capture after ambient cars began advancing from a stored heading and accepted road movement, using the recovered 15–30 m and 12 m lookahead samples. The sampled cars are upright and remain on the road in this view. The gray body paint, nearly black wheels, broad empty pavement, and overall scene still look unlike retail gameplay; speed and steering-force response remain approximations.
Open or save full-size capture · Read the current traffic and texture findings
Chrome capture after removing the viewer behavior that repeated the same route candidate to fill the traffic count. Debug telemetry shows route 101 rendered once; route 102 was rejected because its preview pose overlapped route 101 at 0.003 m. The visible ambient car sits on the roadway in this bend, while the scene still has dark lighting and the sampled routes do not reproduce retail lane changes, intersections, or avoidance actions.
Open or save full-size capture · Read the roster and action/controller evidence
At route 1095, distance 600, the source and dist pages both report a ready scene and four rendered cars on separate static routes. Motion debug reports no cross-route transition targets. The retail `0x00420ed0` path selects a route phase and commits it only after a live FuturePoint queue update; those per-car candidates are absent from the static export, so the viewer does not choose a guessed junction branch. As its lookahead passes the route fragment end, movement holds at the last reachable sampled pose; a newly seeded car whose requested spawn point is beyond that fragment is rejected as `route-end`. This capture still shows the scene's dark look and a car on the left-side paved branch; lane, intersection, and avoidance behavior remain unresolved.
Open or save full-size capture · Read the exact selector and queue path
The far-left car from r157 was at world position (22.667, −2006.629, 11.612); the road-triangle sampler returned no support within 4 m, despite the adjacent raised surface looking paved. Traffic now rejects unsupported spawn poses and refuses to initialize a driver off the road mesh. In this capture the remaining car on route 1026 is supported at Z 11.525, and its frame-up Z is 0.99976, so it is upright. Other route fragments still fail this conservative support check and are skipped.
Open or save the support-check capture · Open the checked view
The r159 seed for route 1134 landed just past its supported road section. A 2.5 m scan found supported pavement earlier on that same decoded fragment, so the viewer now chooses a nearby supported point that falls within the forward view corridor. Route 1067 also contributes a supported phase; route 1026 remains supported. The three selected route keys are distinct (1134, 1026, 1067). Runtime telemetry reports a road-triangle hit at each pose and frame-up Z values from 0.99976 to 1.0. Movement still stops when each fragment ends; no inferred junction joins were added.
Open or save the three-car capture · Open the checked view
After rebuilding the network package, Chrome loaded the L4RB city from its 12 MiB GLB chunks. The page reached “City scene ready” with 19,068 nearby triangles, 110 nearby textures, 99 traffic routes, and 12 ambient body models. This confirms the packaged chunk manifest works beyond L4RA; the scene remains very dark and traffic is still a route-fragment preview.
Open or save the L4RB capture · Open the packaged L4RB scene
The opaque RDP shader had stopped using the source vertex colors. This capture restores the previously audited world operation: texture RGB multiplied by COLOR_0 RGB × 2, clamped to 1.6. The authored lane textures and UVs provide the markings. Chrome finished all 110 nearby texture uploads without losing its context. The broad gray road is now dark prelit pavement; the scene still needs lighting work. Only two local fixtures are active here, and this L4RB asset supplies no selected environment texture for vehicle reflections. This is a material correction, not evidence of a complete retail lighting match.
Open or save the prelight capture · Read the audited prelight operation
The ambient export binds the available shared CARBOTTOM, DRIVER, WINDOW_TRAFFIC, and TIREBACK images. The flat body paint is the unresolved material hash 0xd6d6080a. The selected 240SX has the same issue: its decoded stock GLB contains three image textures, while body paint remains runtime state rather than a paint image. I checked all 15 ambient folders and indexed every nonempty installed car texture and vinyl catalog; none supplies the paint hash or ambient selector key. No replacement paint palette was added.
Open or save the ambient-car capture · Compare the selected 240SX stock-car capture · Read the asset and runtime-state audit
This is the extracted retail RoutesFreeRoam.bin 0x34121 route data over the city, not a WebGL screenshot and not event-race Paths*.bin 0x34148 data. Static 0x34122 endpoint matches and L4RA's 38 0x34123 intersection anchors do not encode each car's active branch: the executable builds its FuturePoint queue and controller state at runtime. The viewer follows these free-roam route samples, but the speed, lane choice, junction decisions, and vehicle response remain approximate. A live XP memory snapshot is the next evidence needed before changing that behavior safely.
The viewer now applies the OpenUG2 reference renderer's world MODULATE2X vertex-color operation to city materials while preserving raw vertex colors for vehicle draws. This fully loaded L4RB capture shows brighter buildings and foliage; the player car is unchanged and the scene remains dark overall. The reference implementation supports this material operation but is not proof of an exact retail match.
Open or save the packaged Chrome capture · Read the source comparison and limitation
This packaged Chrome capture uses the same city prelight change in L4RA at route 1095, distance 600. The page reports 23,606 nearby triangles, 832 routes, and 12 ambient models; all 114 nearby city textures loaded and WebGL context remained available. The road, signs, and street fixtures are visible, while the player starts over the center marking and the scene remains darker than retail gameplay.
Open or save the L4RA capture · Read the material comparison
The measured stripe sample at the default route-450 start scored 0.400, below the old detection threshold. The viewer now accepts it and applies the route's authored 2.148 m right-side corridor, whose sampled texture has no stripe and whose position is road-supported. A 2.2 second W input moved the 240SX about 34 m to 29 mph; the screenshot shows three traffic cars ahead and no road-failure state. This lane placement remains a preview approximation of the retail AI.
Open or save the packaged driving capture · Read the probe values and limitation
The L4RA GLB's authored sky environment image is now retained when nearby city blocks change. After driving through a stream boundary, Chrome reported 27,352 nearby triangles, 138 textures, an intact WebGL context, and source environment texture 1459 loaded for the 240SX. This protects the recovered reflection input; the renderer's reflection match and L4RB's missing sky material remain open.
Open or save the stream-boundary capture · Read the resource lifetime audit
The viewer reads retail free-roam street routes, not event racing paths, but its cars still use a static route follower and lack the game's live junction choices, speed response, and obstacle behavior. The available ambient textures bind correctly; body paint comes from a runtime material state absent from the extracted texture catalogs. The audit lists the exact decompiled calls and XP memory fields needed to recover both.
View the ambient-car capture · View the extracted free-roam route map · Read the retail-versus-viewer trace and XP capture plan
The L4RA light export contains one broad EL_NightA_Moon record; L4RB repeats the same record three times. The renderer now deduplicates these authored records and uses their inner radius as the full-intensity range. Chrome loaded all 114 nearby textures with five fixture lights and one global light active. The scene remains dark compared with retail gameplay, and the intensity/equation are still approximations.
This close Chrome frame forces the retail COUPE BODY_A and confirms its body shape, window surface, upright orientation, and road-supported placement. The exported underbody, window, driver, and tire maps load; no body-paint image resolves for material hash 0xd6d6080a. The gray material factor is only an exporter fallback, and the tire surfaces remain too dark. I have not substituted guessed paint.
Fresh 960 × 540 capture from the packaged viewer in headless Chrome with WebGL2. The city upload completed all 103 selected textures and the traffic fleet is present; the page reports no error and the route audit reports 832 free-roam routes with no synthetic road markings. The close ambient car still has a flat fallback body finish and nearly black tires. Its body paint depends on a runtime material state that is absent from the extracted files, and the XP VM remains at the login screen, so the retail paint state has not been captured. This screenshot shows the current result, not a completed export.
Open or save the full city capture · Read the traffic and paint evidence
Captured from the packaged viewer through headless Chrome. The city loads and ambient cars are upright on the roadway, but their body panels appear plain white because the extracted files contain no ambient paint images and the retail color comes from a runtime-selected material state. This update removed the guessed blue-gray body tint and synthetic paint finish while preserving the embedded window, tire, and driver maps; it does not recover the game's paint colors. The HUD still shows traffic assets loading at 7/12 in this frame.
Open or save the full-size capture · Read the source asset and runtime capture evidence
Headless Chrome reports “City scene ready,” 103 nearby textures, all 12 quota-eligible traffic body models, and five rendered traffic poses. The capture shows an upright ambient car ahead on the road; its body remains plain white because retail paint state is unresolved. This verifies asset loading, not a faithful retail traffic controller or completed city match.
Open or save the full-size capture · Read the traffic controller comparison
During a short W-key drive, the player advanced about 7.5 m on the road and stopped at an ambient traffic collision; road support remained valid and no building hit was reported. This capture also exposes the unresolved white body fallback on both traffic cars. It demonstrates basic movement and collision detection, not retail traffic driving or recovered paint.
Open or save the full-size capture · Read the traffic controller and paint gaps
Fresh headless Chrome capture from the served dist build. The 240SX and ambient traffic are present on an authored marked road, and the page reports ongoing city texture uploads. Large overhead geometry dominates this view, so this is evidence of the current visual problem, not an acceptance capture. The screenshot is also early enough in the upload that distant scenery textures remain incomplete.
Open or save the full-size capture · Read the current traffic and paint evidence
Fresh capture from the same packaged Chrome build after all 114 nearby city textures uploaded; WebGL context remained intact. A 1.8 second W input moved the 240SX to 16 mph in first gear, with the original engine recording active and no road-support failure. Once the textures finish, the large white geometry seen in entry 246 resolves into the street scene. Ambient bodies are still plain white, and this local drive check does not establish citywide placement or retail AI fidelity.
Open or save the full-size driving capture · Compare the earlier in-progress load · Read the remaining traffic and paint gaps
The route basis places Route 1055 at an authored RDP_GRAVEL_A_CK surface at distance 78. The viewer previously excluded all gravel from road support, which left this route start undrivable. The packaged build now includes that road material: with all 167 nearby textures uploaded, a 2.2 second W input reached 24 mph in first gear, with no road-support, building, or traffic collision failure. The gravel image comes from the exported material. This verifies the tested approach, not every route or city placement.
Route 1014's tested point lies on the authored RDP_SHOULDER2GRAVEL surface, which the viewer had excluded from road support. The packaged build now accepts that source surface. With all 164 nearby textures uploaded, a 1.8 second W input reached 23 mph in first gear; WebGL stayed intact and no road-support, building, or traffic failure was reported. The capture shows the car crossing from the gravel shoulder onto the source-marked road.
The 240SX stock export uses black vertex-color masks on body panels whose actual color is selected by a retail runtime material state. The WebGL preview now keeps those masks from rendering the body black and uses a neutral gray fallback while leaving real decal, window, brake, and wheel texture references intact. The screenshot confirms the garage model loads with no browser errors. It is not an authentic paint texture or recovered retail color; the missing class-0x3f paint descriptor and selected runtime state still block those.
Open or save the full-size capture · Read the car material evidence · Read the ambient traffic logic gap
The packaged dist build loaded the Route 1014 city scene with 164 nearby textures, 67,387 nearby triangles, and no browser errors. The 240SX body now uses the neutral unresolved-paint fallback. This capture also shows that the assembled car still has no visible wheels in this view, so the vehicle appearance is not finished. No retail body-paint texture or selected paint color has been recovered.
Open or save the full-size capture · Open this route · Read the paint material evidence
After excluding the stock car’s black runtime vertex masks from RGB modulation, the headless Chrome side view shows the assembled wheels, rim detail, and glass maps. The body uses a neutral gray fallback because the selected retail paint state is unavailable. The rear chase view in entry 251 concealed the wheels; this side capture confirms they are present in the assembled car.
Open or save the full-size capture · Compare the packaged city view · Read the material evidence
Fresh 1280 × 720 capture from the packaged viewer at free-roam Route 450, distance 300. All 96 nearby city textures were uploaded; the screenshot shows the authored crosswalk, lane paint, roadway, streetlights, surrounding buildings, and traffic at the intersection. A debug scan found 28 nearby XB building groups but no XB/RDP road-triangle intersections in this 120-unit crop; this does not assess other object families, overhangs, or other streets. The ambient-car roster was still finishing its last model, and player/traffic bodies still use neutral paint because the retail runtime paint state is not in the extracted files. This is a loaded scene checkpoint, not a claim that the export is complete.
Open or save the full-size capture · Open this packaged viewpoint · Read the local placement audit · Read the traffic and paint evidence
A 1.8-second W input from Route 450 distance 510 moved the 240SX to 19 mph in first gear; the original engine audio was active and the browser reported no errors. The frame shows the car on the marked roadway with the double yellow center line and white dashed lane edges from the extracted road textures. Four of 120 nearby textures were still uploading, so this documents a short straight drive rather than complete citywide driving behavior.
Open or save the full-size capture · Open this lane start · Read the remaining traffic and paint gaps
The packaged viewer now follows at a lower angle and target height. In this fresh capture, a 1.8-second W input reached 21 mph in first gear; all four wheel positions are visible and the source double yellow line and white lane dashes remain in view. The browser reported no errors. A debug scan found 17 nearby XB building groups but no XB/RDP road-triangle intersections in the 120-unit crop. Four of 120 nearby textures were still uploading; the body remains neutral because the retail paint state is still missing.
Open or save the full-size capture · Open this lane start · Read the local placement audit · Read the paint source evidence
The region now frames its actual scene for orbit inspection. It contains a small circuit layout and has no recovered free-roam routes, so drive mode is disabled; an earlier pavement-only fallback placed the car on an empty slab. Chrome loaded 29,089 nearby triangles and four textures with no errors.
Open or save the full-size capture · Open L4RH
L4RR also has no recovered free-roam routes. The viewer frames its sparse source scene for inspection and disables drive mode; Chrome loaded 16,404 nearby triangles and two textures with no errors.
Open or save the full-size capture · Open L4RR
Fresh headless Chrome capture of the served dist build. The viewer reached “City scene ready” with 90,091 nearby triangles, 221 nearby textures, and 12 ambient body models loaded; no browser console or request errors were reported. The road and source markings are visible, with ambient cars ahead. The player body is still neutral white because its retail paint state is unresolved, and this single viewpoint does not validate placement across the whole city.
Fresh capture from the packaged build after it reached “City scene ready.” A 1.8-second W input moved the 240SX to 17 mph; the HUD reports 3,147 RPM and the original engine recording active. The car is offset into the marked lane, source road markings and ambient cars are visible, and the browser reported no console, request, or road-support errors. The body remains neutral white because the retail paint state is unresolved; this short drive does not prove citywide placement or retail traffic AI.
Open or save the full-size capture · Open the packaged city viewer · Read the traffic and paint gaps
The viewer now applies its existing clearcoat lighting to unresolved paint surfaces while keeping the body color neutral. This packaged Chrome capture shows a soft city-panorama highlight on the body; the 240SX reached 19 mph and its original engine recording was active. The browser reached “City scene ready” with no console, request, or road-support errors. This improves the surface response but does not recover the retail paint color or finish parameters; compare the matte r181 view in entry 259.
Open or save the full-size capture · Compare the earlier neutral-matte body · Read the paint source evidence
Fresh packaged Chrome capture of a second city region. L4RD loaded from its split city manifest with 22,683 nearby triangles, 108 textures, and 26 free-roam routes. A 1.5-second W input reached 22 mph with original engine audio active; no browser request/console errors or road-support failure were reported. The view shows the selected street beneath an elevated road and an ambient car ahead. This verifies that region’s loading and a short drive, not full-city placement or a retail gameplay match.
Open or save the full-size capture · Open L4RD in the packaged viewer · Read the current placement audit
The packaged city viewer exposed 29 stock car choices; this capture follows selecting the 350Z and driving it on the default L4RA lane. The HUD reports 21 mph, 3,834 RPM, and original 350Z engine audio active. Runtime diagnostics show the 350Z Ginsu mapping loaded and AudioWorklet running; the browser reported no request or console errors and no road-support failure. Its body pigment remains the neutral unresolved fallback.
Open or save the full-size capture · Open the packaged city viewer · Read the paint source evidence
The 350Z remains selectable from the 29-car roster and drives on the authored L4RA lane. After review of the first capture, unresolved paint now receives a 38% strength clearcoat to avoid clipped highlights; the body color remains the neutral fallback. The HUD reports 20 mph, 3,609 RPM, and original engine audio active. Runtime diagnostics show AudioWorklet running with the mapped 350Z Ginsu recording; no browser errors or road-support failure occurred. Retail paint color remains unresolved.
Open or save the full-size capture · Compare the first clearcoat pass · Read the paint source evidence
The browser now applies the executable's ordinary throttle-output limits before its approximate acceleration model: rising at no more than 1.0 per second and falling at no more than 1.5 per second. After a 1.8-second W input and release, the 240SX reached 27 mph; the original 240SX engine recording remained active during the drive. Chrome reported no console/request errors or road-support failure. This recovers the throttle command response stage, not the game's force, engine, or drivetrain simulation.
Open or save the full-size capture · Open the packaged city viewer · Read the remaining driving-model gaps
Chrome 150 rendered the source L4RA scene at 800 × 600: 90,091 nearby triangles, 832 free-roam routes, and all 221 selected city textures uploaded with no browser errors or WebGL context loss. Eight of twelve ambient body models had loaded when this frame was captured. The 240SX remains in the neutral body-paint fallback because its selected retail paint state is runtime data not present in the static exports.
Open or save the full-size capture · Open the packaged city viewer · Read the traffic and texture audit
In the packaged WebGL build, holding W for 1.6 seconds moved the 240SX to 12.71 m/s (28 mph). All 114 selected nearby city textures and all 12 quota-eligible ambient models had loaded; six traffic cars were rendered. The HUD reported the original 240SX engine recording active, and Chrome 150 reported no page errors. The capture uses a 120-unit city radius. The body remains on the neutral paint fallback because the selected retail paint state is unavailable in the static export.
Open or save the full-size capture · Open the packaged city viewer · Read the driving model gaps
The optional barrier toggle now uploads the extracted XY boundary segments after WebGL initialization and draws the diagnostic lines over the scene; the source records have no Z coordinates. Chrome 150 rendered the loaded L4RA scene with the overlay enabled. The normal drive view remains unchanged when the toggle is off. The 240SX body still uses the neutral fallback because the retail paint selection is runtime state.
Open or save the full-size capture · Open the packaged city viewer · Read the traffic and material evidence
Fresh Chrome 150 capture from the normal /webgl-city/ URL. That page now reads the compressed 15-part city asset from the package and reached “City scene ready” with 221/221 selected city textures uploaded and no WebGL context loss. The road texture and markings are from the extracted scene. The frame still shows the neutral white 240SX paint fallback and remaining visual differences from retail lighting and traffic behavior; this is one loaded viewpoint, not a citywide fidelity check.
Fresh Chrome 150 capture from /webgl-city/ after the page completed loading. Runtime diagnostics report 90,091 nearby triangles, 221/221 city textures, 12 ambient body models, and no browser console or page errors. The scene is drivable and shows the source road markings and traffic ahead. The player's body remains neutral white because retail paint is runtime-selected; this single view does not verify citywide placement or retail traffic steering.
Captured from the loaded L4RA city in Chrome 150 with drive mode disabled. The overhead view shows the road and sidewalks clear of building footprints at this start; the two oblique views expose nearby buildings, foliage and street props for inspection. These sampled views do not establish citywide placement. The dark angular street props remain visually suspect and need source asset/material identification before changing their export.
Open overhead view · Open wide oblique view · Open opposite view · Open the workspace viewer
Drive view on the elevated Bayview road beneath nearby structure.
Open full capture
Drive view through the marked, barrier-lined underpass.
Open full capture
Drive view on the L4RF road course.
Open full capture
Drive view on the L4RG circuit.
Open full captureIn Chrome 150, each selected region loaded and accepted a 1.5-second W input. Player position advanced in all four; speed reached 18–20 mph. No page errors, road-support failures, or building collisions were reported. This verifies short movement from each sampled spawn only; it does not validate full-route continuity or citywide placement. L4RF and L4RG are circuit/track scenes.
Open the workspace viewer · Read the remaining driving-model gaps
Chrome forced three source traffic models onto the recovered L4RA routes and framed each from behind. All three drive upright on the road. Across 15 ambient models and all 39 exported body variants, transformed retail front/rear wheel anchors align with each model's recorded forward axis. Negative route debug values are source point-order comparisons before the viewer reverses waypoint order; they do not mean these cars face backward. Body paint remains the unresolved neutral fallback.
Read the axis audit and limits · Open the workspace viewer
Holding W for 9 seconds moved the player about 166 m from the default start. The viewer rebuilt the neighborhood around the new location: the loaded scene changed from 90,091 to 81,578 triangles and from 221 to 217 nearby textures. Chrome reported no page errors, road-support failure, or building collision; the car slowed after contacting ambient traffic, whose collision response remains approximate. The frame shows the recentered scene and authored road markings.
Open or save the full-size capture · Open the workspace viewer
W moved the car about 11 m at up to 32 mph; the recovered roadway supported the drive.
Open full capture
This stream has airport road-patch triangles but no authored roadway network, so drive mode stays disabled.
Open full captureL4RH has no free-roam traffic route graph, but its exported `TRN_ROADA_CHOP_*` meshes carry the source roadway geometry under the fallback material name `road`. The road index now includes these pieces. Chrome found 4,629 roadway triangles, placed the car on a straight segment with 120 m of supported runway, and moved it about 11 m during a 1.8-second W input with no browser errors or road-support failure. L4RH has no ambient traffic. L4RR contains 20,000 `RDP_AIRPORT_ROADPATCH_A` triangles and remains an orbit-only scene. This adds one limited playable stream; it does not establish a complete Bayview city export.
Open the workspace viewer · Read region and drive notes
Chrome 150 loaded the L4RA scene to “City scene ready” with no page or console errors. The sampled 240 m neighborhood contained 57 candidate building groups and 8,415 building collision triangles; the road/building triangle check found no intersections. This confirms only this loaded neighborhood and camera view; it does not establish correct placement across the whole city. Road markings in the view come from the extracted road materials and textures.
Open or save the full-size capture · Open the workspace viewer
After a 1.8-second W input, the car advanced about 13.6 m. The scene reports 99 recovered routes.
Open full capture
After a 1.8-second W input, the car advanced about 11.7 m. The scene reports 26 recovered routes.
Open full captureChrome 150 reached “City scene ready” in both scenes. The 240SX moved on their extracted road surfaces, authored road materials supplied the markings (zero generated markings), and the HUD reported 27 mph in L4RB and 26 mph in L4RD. No page or console errors were reported. These are separate scene sets in shared world coordinates; this check does not join them into a continuous city drive or validate every street.
Open the workspace viewer
Fresh Chrome 150 capture from route 1013 at 240-unit scene radius. The viewer reached “City scene ready”; its 832 free-roam routes loaded, and the road renderer reported zero generated markings, using authored road materials and textures. No page or console errors occurred. This frame shows a loaded highway view; it does not verify the entire industrial district or compare every placement against retail gameplay.
Open or save the full-size capture · Open this exact route location · Open the workspace viewer
Retail `KIT00_FRONT_WHEE` mesh and its embedded rim texture.
Open full capture
Retail `KIT00_FRONT_WHEE` mesh and its embedded rim texture.
Open full captureThe fixed-width source object names truncate `KIT00_FRONT_WHEEL` to `KIT00_FRONT_WHEE` for these two long model names. The stock exporter now keeps the actual wheel meshes and each car's embedded rim texture; the viewer mounts them at the per-car axle coordinates extracted from `GLOBALB.BUN`. Chrome loaded both stock inspections without page or console errors: 4,869 triangles and 3 textures for the Impreza, 4,857 triangles and 3 textures for the Evo. This removes the previous generic Mustang wheel fallback for these cars.
Inspect the stock Impreza · Inspect the stock Evo · Open the workspace viewer
The packaged index had inherited the source viewer's chunk-manifest path, which doubled the `/dist/` prefix and returned HTTP 404. The source and packaged indexes now point to their respective manifest locations. Fresh Chrome 150 runs loaded both viewers to “City scene ready” with no page, console, or HTTP errors. The packaged capture loaded 90,091 nearby triangles, 221 textures, and 12 ambient traffic models; a 1.8-second W input reached 24 mph with no road-support failure. The 240SX is on the marked lane and the source road paint is visible. The body remains neutral while retail paint state is unavailable.
Open or save the full-size capture · Open this viewer · Open the workspace viewer
Captured from the packaged viewer in Chrome 150 after the scene reached “City scene ready”. The page uploaded all 221 city textures; a 1.8-second W input reached 28 mph, and the debug view showed six rendered traffic cars on supported road surfaces with upright body frames. This confirms movement in the loaded Bayview area. The player body is still a neutral paint fallback, and the visible dark angular props and simplified traffic controller remain open fidelity issues.
Open or save the full-size capture · Open the packaged viewer · Read the texture and traffic evidence
The player steering and heading path now quantizes angles to the 16-bit full-turn representation found in the retail executable. Chrome reached “City scene ready”; W+A moved the car to 14.42 m/s and changed its heading from 2.1561 to 2.3965 radians. This preserves a retail numeric detail, but it does not establish retail-equivalent steering feel: the steering response and tire forces remain approximate.
Open or save the capture · Open the packaged viewer · Read player control decompilation notes
Fresh Chrome 150 screenshot from the packaged viewer at 1280×720 after a 12-second virtual-time budget. It still displays “Loading Nissan 240SX… Reading the exported stock car geometry”, so this capture does not prove the scene became playable. The previous successful r198 capture is still in the feed; this newer run needs a longer deterministic wait or a diagnosed stalled load before the current build can be considered ready.
Open or save the full-size capture · Open the packaged viewer
Captured from the LAN-served packaged viewer in Chrome 150 after all 221 nearby city textures finished loading. A 1.8-second W input moved the player from (-865.21, -442.96) to (-874.85, -428.41) and reached 14.24 m/s (32 mph); browser console and page errors were empty. The car is on the authored road with a traffic car visible ahead. This verifies movement at this spawn only. Paint remains the neutral fallback, and the renderer, traffic behavior, and sampled city geometry still differ from the retail game.
Open or save the full-size capture · Open the packaged viewer
Chrome 150 capture from the packaged viewer after an 8-second W hold from the L4RA start. The 240SX traveled about 139 world units before release and remained on the source-marked road; `__nfsuRoadFailure` and building collision diagnostics stayed clear. The nearby scene re-centered while its remaining 35 textures streamed. This verifies one straight road segment and one scene-stream update, not cornering or citywide coverage.
Open or save the full-size capture · Open the packaged viewer
A 1.5-second W+A input from rest moved the car about 35 world units and rotated its heading by 0.296 rad. It settled at 3.25 m/s with no road-support or building-collision report. The car is close to the left curb in this frame, so this does not establish good handling; the separate high-speed turn check left the road and hit the support guard.
Open or save the full-size capture · Open the packaged viewer · Read driving fidelity notes
Fresh Chrome 150 capture of the LAN-served package after the scene reached “City scene ready” and the nearby texture upload finished. No browser console or page errors occurred. The road, road markings, buildings, foliage, and sky are visible; the 240SX still uses a neutral shader paint fallback, and several dark angular scene pieces remain visible. This is a single L4RA view, not proof of citywide placement or retail visual fidelity.
Open or save the full-size capture · Open the packaged viewer
Fresh Chrome 150 capture at route 450, distance 550, after all 128 nearby textures loaded. The road and its markings come from the extracted stream; traffic cars are visible ahead. The audit found nearby gates and wisteria intersecting road triangles, but this view does not establish that those objects cause the remaining dark angular geometry. The stock 240SX still looks too flat and the city lighting/presentation need work.
Open or save the full-size capture · Open this road segment · Read the current placement audit
Fresh Chrome 150 capture after the viewer reported “City scene ready.” It shows source-marked roadway, placed buildings, and ambient traffic at the default L4RA start; the stream was still loading its last traffic body (11/12) when captured. The scene is drivable, but this frame also shows the unresolved presentation issues: the player car remains flat and untextured, road lighting is too dark, and the camera/car proportions need work. This is a current review image, not a completion claim.
Open or save the full-size capture · Open the packaged viewer · Read driving fidelity notes
Chrome 150 capture with the recovered `240SX_FLAME_184A_MASK` selected over blue paint. The 512×512 mask is decoded from the installed `VINYLS.BIN` and aligns with the stock 240SX body UVs. The optional shop vinyl defaults to None; its white ink color is a preview choice because the retail vinyl-color state is not available.
Open or save the full-size capture · Open the packaged viewer · Read the asset and decoder notes
90,091 nearby triangles, 221 textures, and 12 ambient body models. Dark angular scenery remains visible at the upper edges.
Open full capture
83,611 triangles and the same 221 textures. The comparison narrows part of the dark geometry to the MegaLight family; this diagnostic filter is not applied to normal play.
Open full captureThe normal URL reached “City scene ready” after all nearby textures uploaded. Chrome reported no page or console errors. Hiding the sky dome did not remove the dark shapes; hiding the `XO_M` family did. The nearest source `XO_MEGALIGHTA_1A_00` mesh is about 26 × 26 × 19 world units and its source textures are present, so this is a material/visibility investigation rather than a missing-texture fix. No scenery was removed from the playable scene. See the audit notes.
Open the packaged viewer
90,091 nearby triangles and 221 textures.
Open full capture
88,171 nearby triangles and the same 221 textures. This filter is not used in normal play.
Open full captureChrome 150 reached “City scene ready” in the packaged viewer without page or console errors. Hiding only `XO_MEGALIGHTA_1A_00_0_2` removes the large upper-left dark piece; hiding its shiny-metal or shoebox-texture slices does not. The implicated slice uses the generic untextured `prop` fallback and low source vertex colors. This isolates the exported slice but does not yet identify the correct retail material/light state, so the playable scene keeps the geometry. The unverified city-metal reflection shader was removed after its A/B capture showed no meaningful change. See the audit notes.
Open the packaged viewer
The packaged viewer loaded the rebuilt retail STREAM in Chrome 150 at the normal 240-unit radius: 89,205 nearby triangles, 221 textures, and 12 ambient body models, with no page or console errors. The exporter now preserves source material hashes and unresolved texture keys, and omits the 1,024 valid triangles assigned to the image-less `HEADLIGHTGLOW` slot. The `PAN_*` panorama shells are still excluded because their current material state covers the sky; the two source sky-dome textures remain. The rendering, MegaLight material, distant scenery placement, vehicle paint, and driving model still differ from the retail game. Read the export audit.
Open or save the full-size capture · Open the packaged viewer
23,405 nearby triangles · 114 textures.
19,068 nearby triangles · 112 textures.
19,279 nearby triangles · 68 textures.
22,643 nearby triangles · 109 textures.
39,826 nearby triangles · 92 textures.
28,495 nearby triangles · 53 textures.
29,089 nearby triangles · no free-roam routes.
16,404 nearby triangles · no free-roam routes.
Each rebuilt GLB reached “City scene ready” in Chrome 150 through the LAN-served package, with all nearby textures uploaded and no page or console errors. The route counts show that L4RH and L4RR remain orbit-only. The screenshots also expose unresolved dark geometry and broad pavement in some views; a successful load does not verify retail placement or drivable city coverage. Read the region sweep audit.
Read the full region sweep · Open the packaged viewer
r209 package · 22,643 nearby triangles · 109 textures.
r211 package · 22,445 nearby triangles · 109 textures.
The large black card in the normal L4RD view came from source key 0x2e95ce7d. It has no image in the installed texture packs and its source slices have matid=255. The exporter now omits only ranges with that exact key, removing 1,677 triangles across six regions while keeping the other object slices. The L4RD packaged capture reached “City scene ready” with no browser errors. Other placement, pavement, and lighting differences remain. Read the source-slot audit.
Chrome 150 reached “City scene ready” with all nearby textures uploaded and no page or console errors. Holding W for two seconds and A for one second moved the 240SX from (-865.21, -442.96) to (-879.44, -426.57), reaching 16.41 m/s (37 mph) with a 0.208 rad heading change. Road support and building-collision checks stayed clear. Four ambient traffic cars were rendered on supported pavement with their up vectors aligned to world Z. The car finishes close to the yellow center marking after the turn; this validates movement but does not establish retail handling, whose steering forces remain unrecovered.
Open or save the full-size capture · Open the packaged viewer · Read player-control findings
87,253 nearby triangles · 221 textures · 832 routes · 12 ambient car models. No console or page errors.
Open full capture
The first browser session failed while uploading city geometry batch 1/2. A later run reached the ready scene.
Open full captureA normal Chrome 150 run reached “City scene ready” after uploading all 221 nearby textures, with 87,253 triangles, 832 routes and 12 ambient vehicle models. The earlier angular white shapes were a partial-texture frame at 48/221; they resolve in this fully textured view. No building crosses the road in this one angle, which does not prove placement across the city. The 240SX still looks pale and flat, and a separate cold load lost its WebGL context during city geometry upload. The export is still incomplete.
Open the packaged viewer · Read the region audit
Fresh normal Chrome 150 run on the packaged viewer. Holding W for 1.8 seconds and then W+A for 1 second moved the 240SX from (-865.21, -442.96) to (-886.39, -416.97), reaching 21.01 m/s (47 mph) with a 0.202 rad heading change. Two road triangles were present under the final sample, no road-failure state was reported, and the browser logged no errors. The speedometer and second-gear RPM updated. This confirms basic movement on this segment; it does not reproduce retail handling. The car body still looks pale and flat.
Open or save the full-size capture · Open the packaged viewer · Read driving fidelity notes
The packaged viewer exposes all 29 stock-car choices. Selecting the Nissan Skyline GT-R R34 loaded its exported geometry and wheels, updated the engine audio label, and drove forward at 18 mph with no console errors or road-failure report. The stock body is still rendered in the neutral paint fallback; this confirms the car selection path, not retail paint fidelity.
Open or save the full-size capture · Open the packaged viewer · Read car paint and vinyl asset notes
8.5-second throttle hold · 44 mph · 2 road triangles under the final position · no road failure or browser errors.
Open full capture
After alternating steering inputs, the car ended on the paved apron and road support was lost. It stopped instead of continuing across unsupported ground.
Open full captureThe successful straight run covered about 150 m from the default route-450 start while the viewer streamed nearby city textures; the speedometer and RPM updated and the road stayed supported. A separate mixed-steering sequence ended off the road and triggered the support guard. These captures demonstrate road driving and streaming on this L4RA segment, not reliable turning or citywide coverage. Read the driving fidelity notes.
Open the packaged viewer
The movement guard now checks the four wheel centers from the selected stock car's recovered axle and track anchors. In the same W/A/D sequence, two wheel samples remained on road triangles and two lost support; the car stopped near the pavement edge at 0 mph. This is a more informative contact boundary, but the screenshot still shows the car straddling the sidewalk. Steering remains approximate and this is not a retail handling match.
Open or save the full-size capture · Read the driving evidence
The installed 350Z VINYLS.BIN supplies the selected 350Z_FLAME_701A_MASK P8 texture. Chrome loaded it on the stock 350Z UVs with the original environment map and no browser errors. The blue paint and white ink color are viewer selections. The 350Z catalog contains 853 masks. As of r232, all 29 selectable cars have complete catalogs with 18,671 recovered source masks.
Added a live search field for the recovered vinyl names. Chrome filtered the current 240SX catalog from four masks to the two flame options with no browser errors. This capture verifies the selector interaction; the fleet-wide asset export is still running.
Open or save the full-size capture · Open the packaged viewer
Untextured structures look like large white angular shapes.
Open capture
The white shapes resolve after the city textures upload. The view still shows a pale flat car body and dark road surface.
Open captureThe complete decoded Corolla catalog adds 410 source masks. Chrome selected the Corolla and filtered 35 matching flag masks with no browser errors. The paired captures show the partial-texture artifact disappearing at 114/114 uploads; it was not a city geometry intersection. City appearance remains unfinished: the car paint is neutral and flat, and the road is dark. The source masks are authentic, but saved paint state has not been recovered.
Select the Corolla · Read the city lighting audit
Chrome 150 kept the loading cover active through 114/114 nearby city textures and all 12 ambient vehicle models. The packaged scene reached “City scene ready” with the loading state cleared and no page or console errors. This full-texture view shows the road markings and city blocks without the white partial-load surfaces. The vehicle still has the neutral flat paint fallback, and the overall city and handling remain unfinished.
Open or save the full-size capture · Open the packaged viewer
Chrome 150 loaded the selected Peugeot 106 in the packaged Bayview viewer. The city reached “City scene ready” after 114/114 nearby textures and all 12 ambient models loaded; the page reported no console, page, or failed-request errors. The complete source-vinyl catalog now covers all 29 selectable cars with 18,671 masks. This capture also shows remaining visual work: the Peugeot still uses a pale flat paint fallback and the road surface remains dark.
Open or save the full-size capture · Open the viewer with the Peugeot selected
114 nearby textures · 5 ambient cars rendered · W reached 12 mph.
Open capture
112 nearby textures · 3 ambient cars rendered · W reached 13 mph.
Open capture
68 nearby textures · 1 ambient car rendered · W reached 10 mph.
Open capture
109 nearby textures · 3 ambient cars rendered · W reached 12 mph.
Open capture
92 nearby textures · 2 ambient cars rendered · W reached 11 mph.
Open capture
53 nearby textures · 1 ambient car rendered · W reached 11 mph.
Open captureChrome 150 reached “City scene ready” in all six districts with free-roam route bundles (L4RA, L4RB, L4RC, L4RD, L4RF, L4RG). One second of W input moved the car to 10–13 mph in each, and the sweep logged no page errors, console errors, or failed requests. The selected route counts and visible traffic vary widely. These are separate region starts rather than a seamless district transition; L4RH and L4RR still have no free-roam routes. The screenshots also show the unresolved flat car paint and uneven road lighting, so this verifies coverage and movement, not city fidelity.
Open L4RA · Open L4RB · Open L4RC · Open L4RD · Open L4RF · Open L4RG
29,089 nearby triangles · 6 textures · no free-roam routes or player start.
Open capture
16,404 nearby triangles · 4 textures · no free-roam routes or player start.
Open captureThese earlier captures show L4RH and L4RR before the source-road fallback was connected. At that point each scene reported zero free-roam routes and had no player start. Revision r235 adds starts from the track's `road (missing 120caccb)` meshes and the airport's `RDP_AIRPORT_ROADPATCH_A` surface. They remain isolated road-only scenes without ambient traffic.
Open L4RH · Open L4RR
4,720 road triangles · 120-unit supported runway · W reached 19 mph.
Open capture
20,000 pavement triangles · 120-unit supported runway · W reached 12 mph.
Open captureThe no-route start fallback now recognizes the source road semantics retained in the exported GLBs: `road (missing 120caccb)` on `TRN_ROADA_CHOP_*` track meshes, and `RDP_AIRPORT_ROADPATCH_A` on the airport pavement. Chrome found 120 units of road support in both districts, moved the 240SX under W, and logged no page, console, or request errors. Neither district has free-roam traffic or ambient cars; they remain isolated from Bayview's six route-connected regions.
Drive L4RH · Drive L4RR
Captured from the LAN-served package in Chrome 150 after “City scene ready”: 87,253 nearby triangles, 221 nearby textures, 832 free-roam routes, and 12 ambient body models. The browser reported no page, console, or failed-request errors. This confirms the current L4RA view renders after its roughly 35-second startup load. The screenshot also makes the remaining vehicle problem visible: the default 240SX paint is still pale and too flat. This is a runtime review frame, not a claim of original-game visual fidelity.
Open or save the full-size capture · Open the packaged viewer
This diagnostic was captured 6 seconds after navigation at 16/221 textures. It exposes large unfinished white scenery pieces, so hiding the loading cover this early is not acceptable. Revision r237 restores the full-screen progress cover until the city textures and traffic finish loading. This is an experiment capture, not the current normal view.
Open or save the full-size capture · Open the packaged viewer
Chrome 150 reached “City scene ready” after 221/221 nearby textures and 12 ambient body models loaded. The progress cover remained visible at 16/221 textures and cleared only when the full initial scene was ready; page, console, and request errors were empty. This view confirms the cover prevents the unfinished white scenery shown in item 310. The current 240SX appearance and city rendering still need fidelity work.
Open or save the full-size capture · Open the packaged viewer
The source sky and road are present; remaining textured batches are withheld until their images load.
Open capture
All nearby city textures and the 12 ambient body models are ready.
Open captureChrome 150 revealed the drivable view once the three-quarter texture threshold was crossed (204/221 textures in the captured frame) and reached “City scene ready” at about 30 seconds. Road and recovered sky textures are prioritized, and city batches whose source images are not uploaded remain hidden; this removes the giant white slabs from the earlier 16/221 experiment in item 310. Browser page, console, and failed-request errors were empty. The player car still has its neutral flat fallback paint, and the city rendering still needs fidelity work.
Open the packaged viewer
Fresh Chrome 150 capture from the served r244 build. The scene reached “City scene ready” with 87,253 nearby triangles, 221 textures, 832 routes, and 12 ambient body models. The browser reported no page, console, or failed-request errors. The cube-map sampler is assigned its own texture unit before the first city draw, removing the previous WebGL INVALID_OPERATION; the stock-car exporter now separates a headlight’s fallback from the shared red brake-lamp shader using the source part name. The beige headlamp faces are visible here. The body still reads as flat paint and the city still needs visual-fidelity work; this is a diagnostic capture, not a completion claim.
Open or save the full-size capture · Open the packaged viewer
Freshly loaded player paint with the default color and finish.
Open capture
Same camera and scene, with blue paint selected to make the reflection easier to inspect.
Open captureChrome 150 captures from the served build, from the same front camera. The city’s 512×128 sky panorama now remains the main reflection, with a smaller contribution from the street-level local probe; the gloss clearcoat response is stronger. Both captures reached the ready state without page or console errors. The cloudy reflection is visible on the body and glass, but the paint response is still a WebGL approximation and does not establish a match to retail materials.
Open the packaged viewer
Chrome 150 from the LAN-served build: W for 1.1 seconds, then W+A for 0.8 seconds. The car reached about 35 mph while remaining on supported pavement; the browser logged no page or console errors, and no road-support, building, or traffic-collision diagnostic fired. The car is close to the double-yellow center marking in this frame. This verifies a short drive, not faithful handling; steering and tire response still differ from the retail simulation.
Open or save full-size capture · Open the packaged viewer
Chrome 150 from the LAN-served build. W was held for 1.4 seconds, then A with throttle for 0.9 seconds; the immediate state reached 44 mph with about a 10-degree heading change. This frame was captured while coasting at about 36 mph. The car remains on the marked roadway, and the page and console reported no errors. The motion is still the viewer’s approximate heading-based model, not retail handling.
Open or save the full-size capture · Open the packaged viewer
Chrome 150 capture from the LAN-served r250 build at a close high camera angle. The city’s source cloud panorama is visible across the hood, while the windshield stays darker than the body. The local street probe no longer lowers the panorama coat on front-facing paint. Paint remains an approximation because the retail finish state is not available.
Open or save the full-size capture · Open the packaged viewer
Chrome 150 on the LAN-served build. W was held for 1.4 seconds, then W+A for 0.9 seconds. The immediate state reached 45 mph; this frame was captured while coasting at about 32 mph. The scene stayed loaded with ambient traffic ahead and no page or console errors. Paint reflections are visible in the chase view, though the steering model remains an approximation.
Open or save the full-size capture · Open the packaged viewer
Fresh Chrome 150 capture from the packaged viewer after the loading cover cleared. Holding W for 2.2 seconds moved the 240SX from rest to about 38 mph; the mapped engine recording was active at about 7,000 RPM in first gear. The browser reported no page or console errors. This verifies keyboard drive input and audio startup. Steering and tire response still use the approximate viewer model, and the car's body remains visually flat.
Open or save full-size capture · Open the packaged viewer
Fresh Chrome 150 capture after regenerating the 29 stock car exports. Explicit BRAKELIGHT parts no longer inherit the blue-glass fallback from their shared material hash; their unresolved lens surfaces use the existing red lamp fallback. The 240SX moved to about 38 mph, engine audio was active, and the browser reported no page or console errors. This fixes material classification across 23 car exports. The lens textures are absent from the installed car texture packs, so their exact retail tint and brightness remain unresolved.
Open or save full-size capture · Open the packaged viewer
Fresh Chrome 150 capture from the packaged page at /dist/webgl-city/?debug. After a 3.4-second W hold, the 240SX was moving at 51 mph in second gear; the HUD read 5,098 RPM, matching the viewer's configured 100 RPM/mph for second gear. The earlier 39 mph / roughly 7,000 RPM capture shows first gear, whose viewer ratio is 180 RPM/mph. This resolves that screenshot's apparent gear mismatch, but the RPM ratios and shift thresholds remain hand-authored approximations; retail gear semantics/ratios are still not recovered. The straight drive reached “City scene ready” with 87,253 nearby triangles and 221 textures. Steering, brake, and tire response remain approximate pending free-roam runtime telemetry.
The documented /webgl-city/ route previously requested its city chunk manifest from a source directory that did not contain the packaged chunk files. Its manifest path now points to /dist/reverse-engineering/; the distribution builder still rewrites packaged pages to their self-contained local assets. Chrome 150 loaded the source route to 221/221 city textures at 1280 × 720 with no page errors, console errors, or failed requests. This fixes the alternate page URL; it does not change the handling model or resolve the 240SX's flat-looking body material.
Fresh Chrome 150 capture from the packaged viewer at /dist/webgl-city/?debug. After a 3.2-second W hold the 240SX reached 46 mph in second gear at about 4,640 RPM; five ambient cars were rendered. The city reported “City scene ready” with 221 nearby textures. The screenshot verifies a drivable road and visible traffic, while also showing the unresolved pale, flat player-car body. Steering, braking, and drivetrain ratios remain approximate; this is progress evidence, not a completion claim.
Fresh Chrome 150 capture from the packaged viewer. The selected 240SX's retail 128×64 SHADOWIG alpha mask loaded and is projected under the car; the scene reached “City scene ready” with 221 nearby textures, four traffic cars, and no browser console errors. The shadow mask is extracted source data. Its ground-plane footprint and 0.38 opacity are viewer-side parameters, and the body paint state is still unresolved, so this is a visual improvement rather than a complete vehicle-render match.
Fresh Chrome 150 capture from /dist/webgl-city/?debug. The page reached “City scene ready” with 87,253 nearby triangles, 221 textures, 832 free-roam routes, 12 ambient body models, and no console errors. A short W hold moved the 240SX to 14 mph and started its mapped engine recording. The car is on the road with source lane markings and traffic visible; its paint still looks pale and flat, and the handling and gearing remain approximations. This is a current state check, not a completion claim.
Same-camera baseline from the packaged viewer before the MOLDINGS correction. The neutral reflective finish covers window and body trim, and two rear lamp trim slices inherit the red lamp fallback.
Open or save the full-size capture · Read the material evidence · Open the viewer
The same view after recognizing source hash 0x12c9453c as MOLDINGS. Dark neutral trim now separates the glass and lamp surround. All 29 cars were regenerated: 177 slices changed across 28 cars, while geometry/images and the other 260 lamp slices stayed identical. Exact trim shader values remain unresolved.
Open or save the full-size capture · Read the material evidence · Open the viewer
Settled baseline using the rebuilt trim-corrected GLBs. The default player paint selected full-strength gloss mode 1, bypassing the documented restrained fallback and giving the neutral body a pale peach reflection.
Open or save the full-size capture · Read the material evidence · Open the viewer
The untouched paint now selects the existing 38% clearcoat mode 2. Explicit gloss, metallic, pearlescent and matte selections retain their existing response. The body color, city lighting and glass shader are unchanged. This restores the documented neutral fallback; the selected retail paint state and shader parameters remain unresolved.
Open or save the full-size capture · Read the material evidence · Open the viewer
Fresh Chrome 150 capture of the source viewer after the loading cover cleared; the browser reported no page or console errors. The 240SX is stopped on a marked road, and this camera position shows no building crossing the pavement. The rear glass still reads as opaque and the body looks flat. This stationary frame does not test steering or acceleration, so it is evidence of the current scene appearance rather than completed driving feel.
Open or save the full-size capture · Open this road view · Read the driving fidelity notes
Fresh 1280 × 720 Chrome capture of the packaged WebGL viewer at rest. This view makes the remaining presentation gaps clear: the body and glass read as flat, the pavement is very dark, and the close wall crowds the camera. It is useful as a current-state review, not completion evidence. The handling agent is separately tracing the retail steering and acceleration path.
Open or save the full-size capture · Open the packaged viewer · Read the driving fidelity notes
Same route/camera, with the authored RGB multiplied by packed alpha.
Open full-size capture
The lower gain retains more pavement texture detail beneath the canopy.
Open full-size captureSame packaged Chrome scene at L4RA route 450, distance 7.2. The full-alpha capture raises mean pavement RGB from 175.3 to 195.5 in the same road crop and looks more washed out under the casino canopy. The viewer keeps the earlier 0.7 multiplier by default; source alpha is retained, while the retail diffuse attenuation still needs an XP gameplay comparison.
Open gain 1 view · Open gain 0.7 view · Read the light record note
Chrome 150 capture of the packaged viewer after the loading screen cleared. A 2.5-second W input moved the car from rest to 19.35 m/s; after coasting and a one-second A input the heading changed by 0.181 radians. The car moved on supported pavement and the HUD showed the original 240SX engine recording active. The rear body and glass still look flat, and steering remains the viewer's direct heading approximation rather than retail tire and body physics.
Open or save the full-size capture · Open the packaged drive view · Read the driving fidelity notes
Chrome 150 loaded the packaged standalone stock model with 5,025 triangles, 95 selected part meshes, and three embedded images. The car has separate dark window surfaces and a complete body shell, but its paint remains uniform gray. Source inventory confirms that the body panels have no base-color image; the embedded images are a badge atlas and small wheel, interior, and brake details. The retail paint state and shader parameters are still missing, so this shows the current asset and material gap, not a retail-matched finish.
Open or save the full-size capture · Inspect the packaged 240SX · Read the material evidence
Chrome 150 capture sheet from the packaged driving view. I selected each of the 29 stock models through the in-city car control; all 29 became the active player car, exposed animated wheel meshes, and returned without an asset load error. The source Ginsu map includes a recording for every roster entry. The contact sheet also makes the remaining material gap visible: every untouched car currently uses the same neutral gray paint because the retail runtime paint state has not been recovered.
Open or save the full-size roster sheet · Open the packaged drive view · Read the paint-state evidence
Chrome 150 capture from the packaged page after confirming it loaded viewer.js r261. With ambient traffic off, the previous 7.0 m/s² acceleration reached 58.12 m/s (130 mph) after 8.5 seconds of W; the first A correction hit the road-support boundary. The viewer-only 2.5 m/s² tune reached 20.20 m/s (45 mph) after the same W hold, then completed four 0.7-second W+A/W+D corrections without a road-support stop, ending at 28.80 m/s (64 mph). This improves the sampled city-block drive but is not a retail acceleration value. Steering still uses an approximate direct-yaw model; retail force and controller handoff need XP free-roam telemetry.
Open or save the full-size capture · Open the packaged drive view · Read the handling evidence and limits
Fresh Chrome 150 captures of the packaged build at /dist/webgl-city/. The first screenshot shows the settled drivable view; the second follows a four-second W hold and the HUD reads 23 mph, first gear, and about 4,160 RPM, with the mapped 240SX engine recording active. The capture shows source road markings, nearby city geometry, and vehicles ahead. The page was already driveable while background texture uploads continued; the sampled progress was 192 of 221, and the loading overlay had cleared. The 240SX body still looks pale and flat, and this view does not establish retail-accurate physics or traffic behavior.
Same packaged Chrome 150 drive view at 1280 × 720, before and after reducing the viewer's sky-reflection mix on glass_vehicle materials. The after capture kept the exported dark blue-gray fallback visibly darker while retaining local-city reflections. It loaded all 221 nearby textures and a ready 64 px local environment probe, with no JavaScript errors. The retail executable's WINDSHIELD state contains values 0.3 and 1.0, but their meanings are unresolved; this is a viewer-side visual adjustment, not a recovered retail shader value. No guessed glass texture or transparency was added. The body paint is still a flat fallback.
Fresh packaged Chrome 150 capture after a five-second W input. The HUD reads 27 mph in first gear at 4,920 RPM; traffic vehicles are visible ahead on the marked road. The car advances through the scene and remains on pavement in this straight-line sample. This confirms the current drive view responds to throttle and renders ambient traffic, but it does not establish retail-accurate acceleration, steering, braking, or traffic AI. The handling audit found no further direct player-response change justified by static exports; live XP force and yaw handoffs remain unresolved.
Open or save the full-size capture · Open the packaged viewer · Read the handling evidence and limits
Chrome 150 capture after all 251 nearby textures uploaded and the scene reported ready. A five-second W hold moved the 240SX about 30.9 m and reached 12.26 m/s (27 mph). The traffic diagnostics showed four vehicles, each with road support and an upward frame Z of 0.995–1.000; route 587 advanced about 20.3 m during the same interval. No browser errors were captured. The screenshot shows one traffic car in view, and both visible bodies still have unresolved flat paint. Traffic cruise speeds remain viewer-generated; static exports do not contain the live retail targets.
Open or save the full-size capture · Open the same road view · Read the traffic controller evidence
The packaged WebGL2 scene loaded all 251 nearby textures. A simulated standard Gamepad API reading supplied left stick X = -0.55 and right trigger = 0.45. The recovered retail controller profile produced a steering command of 0.2803; the viewer advanced from 6.33 to 9.42 m/s and changed heading by 0.118 radians over 2.5 seconds. The 240SX stayed on the marked casino road, an ambient car is visible ahead, the original engine recording was active, and Chrome reported no errors. This checks the WebGL controller path with a simulated pad; retail downstream steering and tire physics remain approximate.
Open or save the full-size capture · Open the packaged viewer · Read the recovered controller evidence
Packaged Chrome 150 capture at 1280 × 577 after the city reported ready and uploaded all 251 initial nearby textures. With ambient traffic disabled, an 8.5-second W hold moved the 240SX about 87.8 m and reached 20.93 m/s (46.8 mph). Four 0.7-second W+A/W+D corrections then ended at 31.11 m/s (69.6 mph); the car stayed on supported roadway and Chrome reported no page or console errors. The screenshot shows the car on a marked road with the speedometer, gear, RPM, and engine recording visible. This measures the current viewer tune; acceleration, steering yaw, braking, and tire forces are still not retail-reconstructed.
Open or save the full-size capture · Open the packaged viewer · Read the handling evidence and limits
The packaged Chrome 150 scene loaded all 221 nearby textures with no browser errors or failed requests. RPM now follows each car's stock wheel radius and gear ratios from GLOBALB.BUN. In the 240SX capture, 60 mph registered 6,344 RPM in second gear; shortly afterward the viewer shifted into third at the first source RPM threshold. The frame shows 68 mph, third gear, and the original engine audio active on a marked road. This replaces the shared RPM-per-mph estimate with per-car source gearing. Automatic shift timing and the acceleration, braking, steering, and tire model remain approximate pending a retail gameplay trace.
Open or save the full-size capture · Open the packaged viewer · Read the measured drive data · Read the handling evidence and limits
Chrome 150 rendered the Audi TT from the packaged garage route ?asset=garage&car=TT. The audit compared the garage manifest against the installed ISO's 0x34600 player definitions and geometry packs: all 29 installed player-car models are present as selectable WebGL garage entries, and all 29 have an engine-recording mapping. The screenshot is a standalone stock-model inspection, not a driving capture. The body still uses the neutral paint fallback because retail paint is runtime state; this image shows the current GLB and material appearance rather than a retail-matched finish.
Packaged Chrome 150 capture after all 221 city textures and 12 ambient body models loaded. A 2.5-second W hold moved the 240SX from rest to 5.65 m/s in the game loop; its new engine-curve factor is 1.0 because the 240SX is the reference car. The screenshot was captured after coasting back to a stop. The road markings and two traffic cars are visible, and this camera angle shows no building intersecting the pavement. The player car still has flat white paint. Other cars now use their recovered +0x310 engine curve as a bounded relative multiplier, but the force conversion, clamp and absolute acceleration remain approximations.
Open or save the full-size capture · Open the packaged viewer · Read the handling evidence and limits
Same packaged Chrome session and road after selecting the 350Z in-city. A 2.5-second W hold moved the car from rest to 8.49 m/s; the live game-loop state reported the curve factor rising from 1.50 to 1.55, compared with the 240SX reference factor of 1.0. The capture was taken after coasting back to a stop. This confirms the per-car curve path is active in the running viewer, but does not validate retail acceleration: the normalized comparison and 0.65–1.55 clamp are viewer approximations. The 350Z body remains flat white.
Open or save the full-size capture · Open the packaged viewer · Read the Engine curve mapping
Captured from the packaged viewer after the live Chrome page reached “City scene ready”: 87,253 nearby triangles, all 221 nearby textures, 832 routes, 12 ambient body models, and no lost WebGL context. This confirms the earlier loading screenshots were incomplete captures; Chrome needed more time to finish the embedded texture uploads. In this camera view the road surface and lane markings render and no building visibly crosses the pavement. The road still reads dark, and the 240SX body still uses the neutral flat paint fallback with visibly crude wheel detail; this is not a retail-fidelity claim.
Open or save the full-size capture · Open the packaged viewer · Read the paint-state evidence and missing runtime capture
After a short W input in packaged Chrome, two ambient cars are visibly on the road ahead. The live traffic debug state reported five rendered vehicles; all five had near-vertical world-up vectors and sampled road support in that frame. Their routes still use the reconstructed traffic controller, so this does not establish retail route choice or timing. The bodies retain neutral fallback paint, and the pavement remains dark.
Open or save the full-size capture · Open the packaged viewer · Read the traffic-control evidence and limits
This second L4RA route starts about 1,090 world units from the default area. The live road-material query identifies its surface as the extracted RDP_GRAVEL_A_CK texture. Chrome loaded 28,329 nearby triangles and 95 textures; three visible ambient cars were upright and sampled on supported road triangles. This route is intentionally gravel, not a missing asphalt marking. The capture still shows the neutral-white vehicle fallback and dark lighting; one route does not verify the rest of the city.
Settled Chrome 150 inspection from the packaged ?asset=garage&car=240SX route, zoomed to show the stock materials. The selected model reports 5,025 triangles, 95 part meshes and three embedded images. Its 45 material records include untextured paint_body runs; the embedded images are used by badge, brake, and wheel/interior details, not as body paint. The close-up confirms the pale body fallback and dark opaque glass in this WebGL build. It does not represent the retail paint or windshield shader state, which remain unresolved in the runtime capture notes.
Packaged Chrome 150 loaded all 221 nearby textures, reported the city ready, and logged no page or console errors. A 3.5-second W hold moved the 240SX from rest to 8.78 m/s; a following 1.2-second D hold changed heading by 0.286 radians. The scene shows the car, speedometer, road markings, and nearby traffic during the drive. This confirms keyboard acceleration and heading response in the current viewer. It does not verify retail tire forces or steering feel: the disassembly audit found those depend on free-roam mode and runtime state that the XP activation screen does not expose. The pale body and dark glass are also still visible material gaps.
Open or save the full-size capture · Open the packaged viewer · Read the driving evidence and remaining runtime gaps
Fresh capture from the served /dist/webgl-city/ page after reloading the r268 viewer. Chrome reached “City scene ready” with 101,011 nearby triangles, all 251 nearby textures uploaded, 832 routes, and no page errors or failed requests. The traffic preview selected three distinct routes; no ambient car is clearly visible in this frame. Recovered city light records loaded, and the body-shop navigation prompt is active. The view has no obvious building crossing the road, but pavement color remains too warm and the 240SX body remains a pale fallback. The traffic change removes duplicate vehicle-definition selection; it does not reproduce retail route or AI decisions.
Packaged Chrome capture at the casino route with the default city-light gain. The road's sampled gray texture was being multiplied by warm red vertex color, creating a salmon cast even with city lights disabled. The opaque RDP road pass now uses the source vertex-color luminance as its prelight level and leaves texture hue intact. The pavement reads gray and its yellow/white markings still come from the exported road textures. This is one checked street view; traffic is not obvious in this frame, and the pale 240SX body and dark glass remain unresolved.
Open or save the full-size capture · Open the captured route · Read the road texture provenance
The packaged scene was ready with 251 nearby textures, no page errors, and the default city-light gain. Holding W for 2.2 seconds moved the car about 7.6 world units and the HUD reached 12 mph with the original engine recording active. Traffic diagnostics reported three upright, road-supported rendered models, but none is clearly visible in this camera frame. This confirms motion and the pavement change; retail steering forces, traffic decisions and clear player-car paint remain open.
Open or save the full-size capture · Open the captured route
Packaged Chrome capture after raising the drive camera pitch from 0.12 to 0.24 radians. Two ambient cars are now clearly visible on the road ahead in this view; diagnostics report both rendered models upright and supported by road triangles. The player reached 17 mph, the city finished loading with no page errors, and the road remains gray with source-textured markings. Traffic motion and steering are still preview models, and the player 240SX still has unresolved paint and glass materials.
Open or save the full-size capture · Open the captured route · Read the driving fidelity gaps
Packaged Chrome loaded the city with no page or console errors. A short throttle input moved the 240SX to 10 mph; the capture shows the textured lane markings and upright ambient traffic ahead. The code now uses each visible traffic car's lane-adjusted position for player collision checks as well as rendering, so the hit point follows the displayed vehicle. This capture verifies the running scene and movement, not collision response or retail traffic steering. The player body paint remains the neutral fallback.
Open or save the full-size capture · Open the captured route · Read the traffic reconstruction evidence
Packaged Chrome loaded the route with no page or console errors. A short throttle input moved the player car to 12 mph; the traffic diagnostics report upright, road-supported models on the visible road. The traffic selector now treats each nonzero density quota as eligibility and chooses evenly among eligible model definitions; quota bytes are not used as selection weights. This follows the recovered manager assignment path, while live candidate counts and PRNG state remain unavailable. The player paint is still a neutral fallback.
Open or save the full-size capture · Open the captured route · Read the recovered traffic manager path
Packaged Chrome loaded the L4RD scene with 63,884 nearby triangles, 155 textures, 26 free-roam routes, and no page errors. A three-second W input moved the 240SX about 15 world units to 11 mph. Two traffic models were rendered upright on the road. Their motion is still the reconstructed route follower. This second-region capture confirms the scene streams and is driveable on its authored pavement; the industrial street is visibly dark, and citywide layout and lighting remain under review.
Open or save the full-size capture · Open L4RD in the packaged viewer · Read the open city placement audit
These packaged Chrome captures use the same stopped-car camera. Raising city-light gain from 0.7 to 1.5 lifts the sampled road crop mean from RGB (40, 41, 44) to (48, 49, 55), while the elevated ramp brightens more. The two nearest authored fixtures are 37 and 50 units from the car, beyond their 30-unit radius, so they add no light at this spot. The shader already applies the recovered type-2 radial falloff; a global multiplier does not fix this local gap. See the detailed measurement and limits.
Open gain 0.7 · Open gain 1.5 · Read the L4RD lighting audit
The running city page has 29 stock player-car entries. Selecting RX-7 replaces the driving body and changes the engine label to the original Mazda RX-7 recording while keeping the L4RD road scene loaded. This verifies a second stock car can be selected in the driving view; it does not verify every car individually. The pale body, dark industrial lighting, traffic fidelity, and retail driving feel still need work.
Open or save the full-size capture · Open L4RD driving view · Read the recovered retail lighting profile data
Same-camera packaged Chrome captures at the normal L4RD light gain. The uncustomized car now starts with the existing gloss preview (shader strength 1.0) instead of the restrained unresolved-material fallback (0.38). The roof and body reflections read more clearly, but the sky and ramp produce broad warm bands; this is still a WebGL preview rather than a recovered retail paint setting. Existing saved finishes remain per car. City lighting and player materials remain under review.
Open gloss view · Open restrained view · Read the comparison notes
Packaged Chrome reached “City scene ready” with 63,884 nearby triangles, 155 textures, and 26 routes. A 2.5-second W input reached 13 mph; the road-support failure diagnostic remained clear and the page had no error. This single straight-road view shows no building intersecting the pavement, but the street remains dark and the car has broad warm reflection bands. It does not verify the full region.
Open or save the full-size capture · Open the L4RD driving view · Read the visual audit
The packaged Chrome page reloaded with no visible scene error. Traffic debug state reported three rendered cars, each upright (world-up Z above 0.9999) and supported by road geometry. The traffic selector now allows a definition to fill its recovered density quota instead of limiting it to one car. Only one small traffic car is obvious in this camera; the screenshot does not verify citywide traffic behavior. The player paint still has broad warm bands, and the road is dark.
Open or save the full-size capture · Open L4RD driving view · Read the quota change and limits
Same packaged L4RA route point before and after. The source road texture scored 0.290 on the center stripe, below the old 0.35 detection threshold; the valid authored corridor moves the car 2.129 m to a sample scoring 0.023. The corrected capture shows the 240SX in the lane and clear of the double yellow. The road triangle still supports the car, and seven traffic cars report upright, road-supported poses. This fixes the measured Route 450 point only; corridor selection on other roads remains under review.
Open the before capture · Open the corrected capture · Open the corrected route point · Read the texture and corridor evidence
An 8-second throttle and brief steering sequence moved the 240SX about 78 world units from the corrected Route 450 start into the casino intersection. The packaged viewer streamed a new neighborhood and reported 26,712 triangles, 140 textures, and no road-support or building failure. Traffic contact briefly stopped the car; a short steering and throttle input moved it again to 7 mph. Four ambient cars reported road-supported, upright poses in the final frame. This checks one route and one stream update; traffic decisions and collision response remain approximations. The road is dark and player paint remains pale.
Open or save the full-size capture · Open the tested drive start · Read the capture details and limits
The car stopped at a junction edge when only two of its four wheel anchors found road triangles. Reversing moved it clear of the curb; throttle then returned it to the intersection. The final drive state was on pavement, with six ambient cars upright and the city still ready. The support rejection is intentional at this location; it does not prove every city curb or intersection is covered.
Open or save the full-size capture · Open the tested drive start · Read the drive telemetry and limits
Packaged Chrome loaded viewer r282 and reached the ready city. Holding S for 1.8 seconds moved the 240SX backward to 8.63 m/s; after release, the HUD showed reverse at 16 mph and 3,353 RPM. There were no page errors. This verifies the viewer input path only: retail registers brake and reverse as separate events, and the live reverse-to-drivetrain path remains unverified. The reverse curve-sampling correction uses each car's exported reverse ratio, final drive, and wheel radius; its acceleration remains approximate.
Open or save the full-size capture · Open the tested drive start · Read the driving evidence and remaining gaps
The retail material hash 0x0fedee40 identifies DULLPLASTIC, so the car exporter now keeps that surface out of the player-paint and clearcoat path. Chrome 150 loaded the packaged r283 scene with no browser errors. The body still appears pale and the windows remain dark; this capture makes that remaining appearance gap visible rather than presenting the semantic correction as a full paint or glass fix.
Packaged Chrome reached “City scene ready” with WebGL2, a local environment cube, and five active nearby city-light sources. No page errors appeared. The shader now handles the signed radius used by authored city lights and adds paint/glass reflection RGB without replacing the base pigment. The body still looks too pale and the windows remain opaque; glass alpha and coefficients are not recovered, so this is an intermediate capture rather than a completed car-appearance fix. The catalog audit found no corresponding paint or glass textures in the 31 installed car catalogs. A later packaged Chrome roster check loaded all 29 selectable stock cars without runtime errors and restored the 240SX; that confirms loadability, not visual or handling fidelity for each car.
Open or save the full-size capture · Open the tested drive start · Read the renderer change and limits · Read the texture catalog audit
All 29 packaged stock selections loaded in Chrome with their shadow textures and no runtime or console errors; the viewer was restored to the 240SX. The contact sheet shows upright cars with distinct silhouettes. Their paint is still pale and largely untextured, and their windows remain opaque. This is a load and rear-view pose check, not proof of per-car handling or full visual fidelity.
Open contact sheet · Scroll through all 29 full-size car captures · Read audit limits
Live packaged Chrome capture from the Route 450 start. A 2.9-second W input followed by 0.7 seconds of W+A moved the 240SX about 19.7 m; the debug state ended at 7.49 m/s with a 0.18-radian heading change. The car crossed onto the double-yellow marking during this short turn, so this is direct evidence that the current response does not keep the car centered through a steering input. The page remained loaded and the original 240SX engine recording was active. This does not identify a correct retail yaw or tire-force value; those still need the free-roam runtime handoff documented in the driving audit.
Open or save the full-size capture · Open the tested drive start · Read the retail driving evidence and remaining capture gap · Read the capture measurements
After restoring the missing 240SX reference powertrain row in the source and packaged garage manifests, a packaged 350Z run reported `engineCurveFactor` 1.50 at rest and 1.55 after a 2.4-second W input. The debug speed was 9.04 m/s (20.2 mph), the throttle command was 0.61, and Chrome reported no page exceptions. This confirms that the selected stock curve and 240SX reference now reach the existing bounded multiplier. The multiplier clamp, absolute force conversion, and overall acceleration still need retail gameplay calibration.
Open or save the full-size capture · Open the tested 350Z start · Read the extracted powertrain data and limits · Read the runtime measurements
Chrome 150 reached “City scene ready” after all 120 nearby textures uploaded. The live page reported 24,556 nearby triangles, the L4RA scene within ±120 units, 832 routes, 12 ambient body models, and an intact WebGL context. This rear chase view shows the 240SX upright on a marked multi-lane road with the lane stripes and double-yellow center marking visible. The pavement remains dark, the body paint is pale and flat, and traffic is distant in this stationary frame. A single view does not establish city-wide placement or retail handling fidelity.
Open or save the full-size capture · Open the captured route start · Read the capture details and limits · Read the remaining handling gap
In the same packaged Chrome session, a 2.2-second W hold produced a live speed of 5.86 m/s, first gear, and a throttle command of 0.331. The captured HUD showed 13 mph and 2,402 RPM with the 240SX recording active. The car remained supported by the L4RA road during this straight, low-speed input. This confirms keyboard throttle and movement in the loaded viewer; no steering input or retail acceleration comparison was made. The pale body paint and dark pavement remain visible gaps.
Open or save the full-size capture · Open the captured route start · Read the measurement and limits · Read the remaining retail handling gap
The packaged 240SX draw state now uses finish mode 2, matching the earlier checked r257 default clearcoat behavior, instead of full gloss mode 1. Chrome loaded all 120 nearby textures and reported 24,556 nearby triangles. The body still has no base texture and uses the neutral `#858c94` preview color; this corrects a fallback regression without claiming to recover the retail paint.
Open or save the full-size capture · Open the tested drive view · Read the draw-state evidence and limits · Read the remaining paint and glass gaps
Six live traffic poses on routes 450, 744, 741, 739, 740, and 738 reported world-up Z from 0.99988 to 1.00000 and support from road geometry. The exporter and viewer use per-variant signed axes, and wheel centers come from the matching baked GLOBALB anchors. This capture does not reproduce the user's reported inverted/backward case: traffic is distant, and no route-wide conclusion is possible. The remaining step is a close capture of a failing model/route with its frame and wheel diagnostics.
Open or save the full-size capture · Open the captured view · Read the transform and wheel-anchor evidence
The camera-locked Route 450 sample does not reproduce an inverted/backward car. Close side captures cover all twelve density-eligible BODY_A models; targeted PARCELVAN BODY_B/C captures also remain upright with their per-variant axes. Other models BODY_B/C and other route directions remain unchecked. BODY_C tires look blocky and black at close range, and traffic spacing at the intersection still needs review.
Open or save the close route capture · Scroll through the close model and variant gallery · Read the tested scope and limits
Captured from the currently served Chrome session on the marked Route 450 road. The 240SX is upright and stopped on the pavement; the lane markings are visible. Its body still reads as pale and flat, and the windows read as dark panes. This is the current visual state, not a claim that the car appearance or driving feel is finished. The retail steering trace now proves the 16-bit angle unit and producer branches, but still lacks runtime response values needed to tune yaw safely.
Open or save the full-size capture · Open the captured route · Read the retail driving trace and remaining gaps · Read the paint and glass evidence
The packaged page reached “City scene ready” in Chrome with 24,556 nearby triangles and all 120 nearby textures uploaded. This actual city view shows the 240SX upright on a marked road, with the recovered rear defroster texture visible through the back glass; the nearby road and buildings do not visibly intersect in this frame. The car remains a neutral pale color because the body paint comes from runtime paint state that has not been captured. A single camera view cannot verify other city locations.
Open or save the full-size capture · Open the captured route · Read the source asset mappings and paint limit
A 2.1-second W hold in packaged Chrome moved the 240SX 6.38 m along its heading to 12.2 mph in first gear. The screenshot shows the newly recovered rear-window defroster texture in the actual city drive. This checks straight-line movement and the asset update; it does not establish retail acceleration or steering forces. The neutral body paint and opaque side windows remain visible gaps.
Open or save the full-size capture · Open the tested drive start · Read the live measurements and limits · Read the recovered shared texture mappings
At route 450, distance 550, the prior spawn sat over the double-yellow center marking because the source stripe detector rejected a 0.187 texture score. The authored selector 3 corridor shifts the spawn 2.148 m right, where the same lane texture scores 0.0. The capture shows the 240SX clear of the center marking; a 3-second W hold moved it about 11.8 m to 17 mph while it remained on the roadway. This verifies the measured start only. Traffic lane selection and retail steering behavior still need runtime evidence.
Open or save the full-size capture · Open the corrected route · Read the source samples and drive measurements
Packaged Chrome capture after holding W for two seconds. The 240SX remains the reference car, so its relative mass factor is 1.0; the HUD shows 14 mph in this frame. This checks browser movement and the unchanged baseline, not equivalence to retail acceleration.
Open or save full-size capture · Open the 240SX drive · Read the physics audit
Packaged Chrome capture after a two-second W hold. The Hummer’s exported rigid-body mass scales its existing acceleration estimate to 0.766 at the end of the run; the HUD shows 8 mph in this frame. The test confirms browser integration and relative scaling only. Steering, tire forces, braking, and absolute acceleration remain approximate.
Open or save full-size capture · Open the Hummer drive · Read current driving limits
Capture from the existing Vulkan Chrome session at Route 450, distance 550. The live viewer reports four rendered traffic models on sampled road surfaces and no page error; this view confirms they are present and upright. They remain distant and visually plain, and the static-route preview does not reproduce retail FuturePoint junction choices or traffic-driver physics.
Open or save full-size capture · Open the captured road view · Read the traffic behavior gap
A sampled building triangle shared the road surface height near route 687. The Chrome view at that route point shows the airport structure behind the curb, with no visible building face cutting through the drivable lanes. The mesh intersection is a useful warning, not proof of a visible placement error; this view does not rule out other points on the building.
Open or save full-size capture · Open this route view
A second sampled building/road triangle intersection near route 1013 was checked in Chrome. The captured road shows the structure set behind the roadside barriers; no building is visibly crossing the driving surface in this frame. The road is rendered from the recovered city materials, and this single frame does not prove city-wide placement.
Open or save full-size capture · Open this route view
1280 × 720 Vulkan Chrome capture at Route 450, distance 550, after a 2.6-second W hold. The car moved to 16 mph in first gear; the RPM readout reached 2,919 and the needle moved with it. The XP game bundles supplied the tach face and NOS tick artwork. No page exception was observed. Needle sweep and panel placement remain viewer approximations, and the city lighting, car appearance, and retail handling still need work.
Open or save full-size capture · Open the captured road view
The material named RDP_ROADSTRIP_FREEWAYCE is reused on tunnel ceilings and train structures in the extracted GLBs, so it is no longer treated as road asphalt by name alone. Road compositing is now separated by source mesh role. This Route 450 Chrome capture uses RDP_LANEA_YS_AA_KT, loaded with no page exceptions, and serves as a regression check rather than a lighting improvement. The pavement is still too dark and the car finish remains flat.
The road shader previously ignored RDP material slices on TRN_*_TERRAINA_CHOP_* street meshes. The mesh/material audit found lane, intersection, parking, and road-strip textures in this family across L4RA and L4RD; the renderer now applies the source road composite to those uses while retaining separate batches for non-road uses of shared materials. After a 2.6-second W hold at Route 102, the 240SX moved about 10.8 m and reached 14 mph in first gear at 2,648 RPM. Chrome sampled RDP_LANEA_YS_AA_KT beneath the car with no page errors. This verifies movement and road support there, not retail handling or lighting. The car finish and road shading still need work.
Unedited capture from the existing Vulkan Chrome session at Route 450, distance 550. After a 2.6-second W hold and 1.2-second coast, the HUD reads 11 mph and 2,004 RPM in first gear; the original engine recording is active. The 240SX is on the asphalt with the source double-yellow center marking and lane stripe visible, and ambient traffic is present ahead. This confirms current browser movement and audio start, not retail handling. The paint still looks too flat and the pavement remains dark.
Open or save the full-size capture · Open this drive view
The default city-light gain is now 1.0. In a same-camera Chrome comparison at Route 450 with four local fixtures plus the broad light active, the near asphalt crop increased from mean RGB (59, 60, 60) at gain 0.7 to (66, 66, 67) at gain 1.0; the far-scene crop stayed effectively unchanged. Gain 1.5 brightened the road more but also lifted distant geometry. This capture shows the new packaged default and no page errors. It is a visual preview correction, not a recovered retail exposure value.
Open or save the full-size capture · Read the same-camera measurements · Open this drive view
The all-region material audit found street and airport mesh families missed by the earlier role check. The packaged viewer now applies the road composite to road-marked CITYROAD, runway, TERRAIN03, and other exported source families, while keeping shared materials on unrelated meshes separate. The ready L4RB Chrome capture reports 19,068 nearby triangles, 112 textures, 99 routes, 12 ambient body models, and no page exception. The source lane texture is visible, but the underpass pavement remains too dark; this fixes road-material selection, not city-wide placement or lighting.
The source-role audit found road materials on TERRAIN*_NR/NRND_CHOP, airport terrain, parking-lot, tunnel, road-skid, and drift-marker mesh names. The expanded patterns cover 9,822 RDP road-family uses across the eight exported L4R city GLBs. A packaged L4RA Chrome check found nearby terrain chunks carrying the recovered parking-line material; the classifier accepts them and rejects gravel and tunnel-ceiling uses. The capture still shows dark pavement and the flat 240SX finish. This confirms a broader material-role mapping, not city-wide placement or retail fidelity.
A same-camera L4RB comparison raised the underpass road crop from mean RGB (16.5, 17.1, 17.6) to (26.4, 27.4, 28.1). The nearby building crop stayed at (23.0, 29.3, 20.3), so the lift is limited to classified road materials and leaves the recovered global city-light gain unchanged. Chrome reached “City scene ready” with no page exception. The road still reads as a dark night surface and the 240SX paint remains flat; 1.6 is an image-tuned viewer exposure value, not a recovered retail constant.
Open or save the full-size capture · Read the same-camera measurements · Open the tested L4RB drive
The detector flagged XB_CASINOD_SIDEA_1A_LL_00 at L4RA Route 530, distance 240. Exact triangle clipping removes 44 facade triangles, but the before/after Chrome captures show that this is a covered passage between walls. A 2.8-second W hold moved the car 13.3 world units through it with no road failure or building collision. Hiding the whole group removes 555 nearby triangles and opens a larger hole. Both automatic filters remain disabled because they damage the visible architecture.
The r314 live drive showed the 240SX audio staying at 800 RPM through the first second of full throttle. The r315 viewer uses throttle and a low-speed clutch-slip estimate to raise engine RPM, blending back to the car's stock-ratio wheel RPM by 8 m/s. In the matched Chrome run, RPM rose to 1,549 at 250 ms and 2,812 at 1 s while the car was still below 2 m/s. AudioWorklet was running with the XP-installed 240SX Ginsu recording. The launch blend is a viewer estimate because the XP game remains at login; this does not prove retail clutch behavior.
Open or save the full-size capture · Read the source check and measurements · Open the tested drive and listen
The player renderer draws the blue-gray glass after the opaque body and interior, with 0.76 preview opacity. In this checked Chrome side view the city and interior shapes are visible through the windows, while the recovered environment reflections remain. The original shader's view-dependent alpha is traced, but the live glass texture alpha and blend descriptor are missing, so 0.76 is an appearance choice rather than a retail value. The body paint remains a runtime color fallback.
Open or save the full-size capture · Read the source boundary · Open the checked drive view
The source-mesh color pass maps the black driver-door pixels to MIATA_BASE_A_0_slice2_slot0, an interior_vehicle slice. Suppressing that whole slice removes the black surface but opens a gap along the lower door/body. Enabling back-face culling does not remove it. These are cause-isolation results, not a safe fix. The separate painted DOOR_LEFT_A slice is present. No product material change is justified yet; the car and city are not complete.
The retail XP installation contains a separate Miata right-door object at 0x2fd88, but the old exporter collapsed it onto the left door because their transformed bounds matched. The exporter now keeps explicit left/right parts separate and skips only a stored reflection that folds already mirrored vertices onto the same side. All 29 installed stock car models were regenerated; these clean-room inspection captures sample both Miata sides plus five other cars. The formerly dark Miata door is now assembled from the authored opposite-side door mesh. The neutral gray paint fallback, scene lighting, and retail driving feel remain unresolved; these stills show body geometry, not a finished driving view.
Fresh screenshot from the served /dist/webgl-city/ build. It confirms the rebuilt garage car loads in the city viewer, the marked road and distant traffic render, and the city scene reaches ready state. This rear view does not show the repaired driver-side door. The gray paint fallback, dark road appearance, lighting, and driving feel remain visible areas of work.
Each model was loaded from the packaged viewer and captured in the same diagnostic side view. Every roster entry reached its stock-model-ready state; the sweep shows upright body shells, wheel placement, glass, and opposite-side panels across the full roster. These are clean-room WebGL previews and the same neutral gray paint fallback is visible throughout. This checks the regenerated static models; it does not establish all retail body-kit choices, paint states, lighting, or in-game handling.
Open or save the full-size contact sheet · Open the car inspection viewer
Headless Chrome loaded the packaged L4RA scene with the Miata at rest, then held W for 2.6 seconds and coasted for 0.9 seconds. The HUD reached 11 mph in first gear at 1,629 RPM; the source Miata GIN engine recording was loaded and the AudioWorklet graph was running. Two ambient cars are visible ahead and the lane markings come from the extracted road materials. This verifies WebGL movement, HUD response and audio-path startup; it does not prove the audible blend or handling matches retail. The gray paint and night lighting still need work.
Open or save the full-size capture · Open this tested drive
The Celica viewer was held at Route 450, distance 630, over RDP_INTERSECTMIDTILE; both captures use cityRoadGain=1. Only cityLightGain changes, from 0 to 0.4. The higher setting adds visible fixture response while keeping the intersection texture readable; the packaged default gain of 1.0 washed this crossing out in a separate capture. This isolates a WebGL light contribution, but it is not an XP lighting comparison, so I have not changed the default.
Packaged Chrome capture at L4RA Route 450, distance 700. The default preview now uses road gain 1.0 and city-light gain 0.4. Compared with the same-pose 1.6 road-gain frame, pavement texture contrast returns while fixture lighting remains visible. Several traffic cars are visible ahead. A shared 4.2 m cutoff also keeps traffic rendered until the existing player-contact check can act, avoiding a hidden-car collision. The crossing still lacks a clear center marking in this angle; the Celica paint and retail traffic behavior remain unresolved.
Open or save the full-size capture · Read the comparison and limits · Open this city view
In the packaged Chrome drive, a held-W run reached 13.1 m/s before the viewer's traffic contact guard slowed the car; the pavement support check stayed clear. The capture shows traffic still visible ahead after releasing W. This checks the fixed viewer visibility/contact mismatch, not retail collision geometry or traffic reaction.
Open or save the full-size capture · Read the measurements and limits
Route 450, distance 550. The recovered windshield alpha path compiles in the served viewer; the window still reads very dark and the body remains a flat silver fallback in this angle.
Open full-size capture
The selected van stays supported by road triangles after removing the duplicated lane shift. Its untextured cargo panel is an unresolved source paint state, not a recovered retail finish.
Open full-size captureThis confirms the traffic pose fix and shader build in the served package. It does not establish retail handling, traffic AI, full glass appearance, or paint state. The pavement still looks dark, the Miata finish is plain, and the forced PARCELVAN remains a white box because its source body paint is unresolved.
Read the asset trace and Route 736 measurements · Open the packaged city view
The pavement is brighter while retaining its authored lane paint. It is still a color-space experiment, not a retail match.
Open full-size capture
This isolated change makes the road and buildings much darker, so texture decoding cannot be changed by itself.
Open full-size captureAll views use Route 450, distance 550, radius 120, and the same Miata. In the sampled road crop, mean RGB is (51,52,52) in the former default, (18,19,19) with sRGB sampling alone, and (73,75,75) when sampling and output encoding are paired. The r330 review promotes the paired pipeline to default because the exported PNG color textures are sRGB encoded; no pixel-for-pixel retail XP comparison is available yet.
Read the measurements and limits · Open the current color pipeline
Captured from the running packaged Chrome page after it reported City scene ready: 66,068 nearby triangles, 188 textures and 832 free-roam routes. The 240SX body remains pale and its rear glass reads as orange stripes. The blue storefront wedge is an authored SHOPBEAM marker at the Mega Low entrance, but its texture key 0x2e95ce7c is missing, so the viewer renders its blue vertex colors as an opaque fallback. Its placement is supported; the original effect state is still unknown. The dashboard is at zero because the car was stationary for the capture.
The orange-brown lines match the embedded REAR_DEFROSTER texture exactly, including its source RGB values and blend material. The main rear-pane texture remains missing, so the dark/opaque window still differs from the game. I found no source-backed opacity or color change to apply.
Exported PNG color textures are decoded to linear for lighting and encoded for display. The paired path brightens the pavement and city while keeping the source lane markings. This live packaged view reached City scene ready with 22,508 nearby triangles, 128 textures, and 832 routes. It is not yet a pixel match to XP; the car paint still falls back to white.
Open or save the live capture · Read the color pipeline note · Open the checked packaged drive
The gray swatch entered the lighting shader too bright, washing the body toward white.
Open full-size capture
The body now uses the selected gray in linear lighting. The road is pixel-identical in the measured crop; this does not restore missing retail paint textures or finish.
Open full-size captureThe default #858c94 swatch was sent to the shader as linear RGB even though the control value is sRGB. It now decodes before lighting. Dynamic reflections also store linear RGB instead of display-encoded values. Matched packaged captures show the bumper crop changing from mean RGB (211, 216, 223) to (151, 159, 169), while the sampled road crop stays unchanged. This fixes the swatch color path, not the missing retail car-body texture or material finish.
Temporary shader diagnostic. The interior bumper pixel is exactly the selected #858c94 swatch; this is not the normal drive appearance.
Temporary diagnostic: the main coat highlight is lower, while the separate lamp-specular branch remains.
Open full-size capture
Temporary diagnostic showing one contribution from the approximate vehicle lighting path.
Open full-size capture
Temporary diagnostic showing another contribution. None of these shader variants changes the real drive view.
Open full-size captureControlled Chrome captures at the same pose isolate the selected pigment and parts of the current preview lighting. The selected pigment returns exactly to #858c94; the normal shader lifts the sampled bumper average from about (133, 140, 148) to (151, 159, 169), with brighter highlights from coat and diffuse terms. No channels clip. These are temporary diagnostics, not improvements to the product renderer, and they do not establish which retail lighting values should replace the preview terms.
Captured from the running packaged Chrome viewer during a two-second W hold: the car moved forward, the speedometer read 9 mph in first gear at 2,518 RPM, the city remained ready, and the page error field was empty. This verifies basic movement and dashboard updates; it does not compare the handling curve with retail XP gameplay.
Open or save the full-size capture · Open this packaged drive view · Read what remains unverified about handling
In the debug state, W+A moved the 240SX and changed its heading from 1.857 to 1.927 radians; the reported front wheel steer angle was 0.156 radians. The scene remained error-free. This confirms input response only, not retail steering calibration. The capture also shows the unresolved opaque blue SHOPBEAM wedge by Mega Low; no transform or transparency guess was applied.
Open or save the full-size capture · Read the measured input trace and limits · Read the SHOPBEAM source audit
Same Route 450 start, camera, car, and viewport; only the authored-light gain changed. At 0.7, the near-road crop rises from mean RGB (75.6, 76.5, 75.8) to (80.9, 81.8, 81.8), while the sampled wall changes by about one RGB level. The source yellow center lines and white lane markings remain visible. This improves the dark pavement slightly; it is still a preview adjustment rather than a retail lighting match.
Open the 0.4 capture · Open the 0.7 capture · Read the measurements and limits · Open the current default
The running packaged Chrome debug page reports one Route 736 ambient van at the road, with an 11 m/s preview speed, 15.46 m lookahead, and road-supported movement. The route source is retail RoutesFreeRoam chunk 0x34121 plus 0x34122 transition links, not event racing paths. Retail's desired-speed handoff needs live candidate/controller values that these static files do not include, and the intersection action also depends on live phase and command state. A follow-up static-graph audit found 29 candidate continuations below the preview tangent threshold; all are same-route/same-waypoint self-transitions, with no distinct sharp turn supported by a unique reciprocal successor. Existing route-basis and signed body-axis checks also found no orientation correction to make. I left traffic behavior unchanged rather than inventing cruise, stop, or turn parameters.
The city uploader previously stopped at four of 128 textures because it waited only for an animation frame, which Chrome may pause in hidden tabs. The loader now has a timer fallback. With the review tab kept in front, a fresh Route 450 viewer progressed to all 128 textures and reached “City scene ready”; the WebGL context remained intact. This frame shows the finished view after the background load completed.
Open or save the ready capture · Read the observed progress and test limits · Open the packaged drive
Read-only offline extraction found seven exact city texture records in GLOBAL/InGameCommon.bun. The Route 736 Mega Low beam now uses the matching alpha-capable SFX_LIGHT_BEAMA texture instead of an opaque blue fallback. L4RA also restores 323 triangles previously omitted for SFX_LIGHT_BEAMB; all eight region manifests now report zero unresolved-black-card triangles omitted. The storefront effect looks materially closer, though its retail blend state and the wider scene presentation still need comparison.
Fresh capture from the rebuilt viewer after changing the city asset URL version so existing browsers cannot reuse the pre-recovery chunks. Chrome reached “City scene ready” with 67,700 nearby triangles, 190 nearby textures, 832 routes and one ambient traffic model; the page error field was empty. The runtime scene query identifies the Mega Low mesh as SFX_LIGHT_BEAMA (0x2e95ce7c). The recovered beam is visible now, while the body finish and overall city lighting still remain far from a retail match.
Matched packaged captures at the same start and camera show almost no additional visible city at radius 400. The viewer loads about 2.5× the triangles and 1.6× the textures, so I kept the current radius. The missing context is split across regional STREAMs: L4RA/L4RD share 92,187 same-name placements at identical full transforms, and 92,109 also match in geometry and material-image payload. Seventy-eight same-transform objects differ and still need handling before a safe merge. The exact-mesh deduplicator also now compares normals and every other authored vertex attribute before it can discard a mesh.
Open radius 220 · Open radius 400 · Read the counts and regional caveat
The centerline start sat over the source double-yellow marking. The decoded route point provides a positive selector-4 corridor offset of 1.8056 m; Chrome confirmed the adjusted pose remains on road triangles and places the 240SX in the right lane. The rebuilt packaged viewer also reached “City scene ready” with an empty error field and the same offset. The viewer applies this only to the exact Route 736 / distance 107 review link when the texture detector misses the stripe; the normal L4RA start is unchanged. A separate W+D run from the farther-right manual 2.15 m position steered into the curb and lost speed after road support failed. That is a viewer boundary stop, not a retail handling match.
Open before · Open source viewer after · Open packaged after · Read the source offset and drive limits · Open this route
The first visual candidate restored nearby overhead scenery at Route 736 without an obvious duplicate road surface or building crossing the pavement in this view. That experiment is now available from the main viewer as world=combined. The more complete L4RB road and drive captures are below. A separate W+D run reached the curb, where the viewer’s road-support guard rapidly reduced speed; retail steering and road-edge response are still unverified.
The main viewer now accepts world=combined and loads the merged eight-region mesh. L4RA and L4RB both reached “City scene ready” with empty page errors; the region selector still chooses each region’s route and traffic data. A short W hold in L4RB moved the 240SX along the raised road while the combined road-support mesh kept it on the surface. This gives the player road geometry across the regional exports, but traffic and route transitions are still region-specific, and the 14 gzip chunks transfer 89.48 MiB and assemble to 163 MiB in the browser. Chrome verified both the initial load and a player move on the L4RB raised road.
The regular compressed L4RA viewer now shows a “Load full Bayview world (89.5 MiB)” link near the top of its controls. Chrome reached “City scene ready” with no error, applied the Route 736 lane offset, and displayed the link without scrolling the controls panel. In combined mode it changes to “Return to L4RA region mesh.” The toggle preserves the current region and route query while switching between the regional and combined scene meshes.
Open the regional-view capture · Open the combined-view capture · Open the combined Bayview driveThe recovered code supports the 4.4 ms physics step, throttle-command slew, controller input curve and packed steering-angle units. It still does not give a usable player steering-to-yaw curve, tire grip, brake force or acceleration scale. The Route 736 selector-4 start is applied in the packaged viewer; the compressed L4RB capture shows road-supported forward movement, while the separate W+D frame shows the current road-edge stop. Those frames verify viewer behavior, not retail handling calibration.
Open packaged lane start · Open supported drive · Open curb stop · Read recovered inputs, stepping, and remaining gaps
This capture is from the live packaged viewer after Chrome reported “City scene ready.” The car still reads as a uniform gray body with dark glazing. The 240SX GLB has no body-paint image or pane image in its material slots; its embedded maps cover badges, mechanical details, underside, and rear defroster. The viewer's body tint and most pane transparency remain explicitly labeled previews until the selected paint and glass state are captured from a running XP game. This is a current-state check, not a completion image.
Open full-size ready-state capture · Read the source-backed texture and runtime-state audit
This packaged L4RA run omitted the single-model trafficModel diagnostic. Chrome reported “City scene ready” with no page error, loaded 12 ambient definitions, and rendered an authored-route taxi on the bend. Its debug pose was upright (frame-up Z 0.999) and supported by a road triangle. A 2.5-second W hold moved the 240SX about 7.3 m to 13 mph while the recorded engine audio played. This confirms a car moves and the default traffic roster loads; only one ambient car is visible in this frame, and retail traffic decision logic remains approximate.
The packaged combined scene loaded all 208 nearby textures with no page error. In Chrome, a 2.2-second W hold moved the 240SX 6.3 m along Route 736 to 11 mph while two ambient vehicles rendered. The screenshots show the source double-yellow road markings and a raised shopping-district view; this viewpoint has no building crossing the pavement. The body paint still uses a flat gray fallback and the engine recording is active, but its retail RPM/load mixing remains unverified.
Open full-size moving capture · Open the same combined drive · Read the retail traffic-driver capture contract · Read the engine audio source and playback audit
I checked 123,639 center and ±1.2 m samples along recovered free-roam routes in L4RB, L4RC, L4RD, L4RF and L4RG. The height comparison uses each route waypoint's recovered Z and tests exported XB_ building triangles from 0.25 m below to 2.2 m above the route. It found zero hits in these sampled corridors. L4RB has 10,823 unsupported road samples and L4RC has 13; those samples still received the building test, but this is not proof that every route segment is drivable. An apparent Old Town hit was an elevated road layer: the actual route is near Z −2.25, while the candidate building surfaces are around Z 14.05. The instance matrix is unchanged between regional and combined GLBs. This does not establish citywide placement or visibility correctness.
After rebuilding dist/ from the current package script, one fresh Puppeteer load reached 208/208 nearby textures in 13.3 seconds, with no page, console, or request errors. In a separate packaged drive capture, a 1.8-second W hold moved the 240SX 4.57 m to 11 mph on Route 736. Its screenshot was taken after the loading panel cleared while the last 8 background textures were still streaming, so the counter shows 200/208. The road and its double-yellow markings are visible. The car body remains the known flat gray runtime-paint fallback, and this check does not validate retail handling or audio mixing.
A fresh packaged drive exposed an over-strict road-support gate: the car stopped when its center and one wheel were still on pavement. The viewer now accepts center plus one compatible wheel, keeping the stricter three-wheel seam fallback. At the road edge, speed now decelerates over time using the existing 12 m/s² preview brake rate instead of multiplying by 0.35 every 4.4 ms step. In the same Route 736 sequence, the car reached the curb, then S backed it about 5 m to 14 mph in reverse. Chrome reported no page or console errors. The screenshot shows the car beside the curb, not proof of a retail curb model or universal recovery. The body remains the known gray paint fallback.
Open full-size capture · Read the measurement and limits · Open Route 736
I changed the actual packaged car selector through all 29 roster entries in Chrome. Every selection completed its loading state and returned to “City scene ready”; there were no page exceptions or console errors. A structural audit also parsed all 29 GLBs and found their triangle counts, materials, and image references consistent with the roster. The screenshot shows the Audi TT selected. This verifies selection and loading only: its neutral body paint and glass still look unlike the retail game, and handling/audio have not been compared for all cars.
Open full-size Audi TT capture · Read the audit scope · Open Route 736
The retail loader trace opens one STREAM<track>.BUN for the selected track; the eight-region union remains an experiment. In a fresh packaged L4RA-only run, W reached 7.49 m/s, A reached 11.32 m/s, steering reached the road edge, and S backed the car about 5.6 m at 12 mph reverse. Chrome reported no page or console errors. The screenshot shows the authored double-yellow road marking and one ambient car. This confirms one route and reverse recovery, not complete city coverage or retail handling. The paint is still flat gray and the glass remains unresolved.
The traffic preview now scores supported free-roam samples by their projected position in the current chase camera and resets the old follower state when a route sample leaves the camera-centered viewing area. The packaged Chrome capture shows one TAXI on the marked road ahead, road-surface-supported and upright. The second candidate is rejected because its sample is behind a building; this uses static route samples and is not the retail traffic controller. Ambient paint remains the neutral fallback, and full traffic behavior is still unresolved.
Open full-size traffic capture · Open the same L4RA drive
In a fresh packaged Chrome run, a 2.2-second W hold moved the 240SX about 7.6 m to 14 mph. One upright, road-supported ambient TAXI remained visible on the authored roadway ahead; the viewer reported no page or console errors. This confirms the current preview can show traffic during player movement. The single car still follows a static-route kinematic approximation, and its body paint is the neutral fallback.
Open full-size moving capture · Open the same L4RA drive
The Windows XP guest lacks a renderable GPU, so this visual check used host Chrome with ANGLE/Vulkan. In the packaged combined scene, the geometry diagnostic found all 8 nearby XO_SPOTSTADIUM_1A_FUN placements intersecting exported road triangles; the screenshot shows the dark structure blocking the skyline. The retail stream trace loads one selected regional STREAM bundle per world-entry call, with no positive evidence that these eight bundles form one simultaneous city. I relabeled the union as experimental and recorded the evidence; this does not fix the city or establish the correct stadium visibility state.
The workspace-only chunk-manifest path had made the packaged L4RA scene fail with HTTP 404. The package-local path is fixed. Host Chrome with ANGLE/Vulkan reached “City scene ready” at 43,607 nearby triangles and 169 textures with no page or console errors. Holding W for 2.2 seconds moved the car 6.4 m; the capture just after release shows 8 mph. A local scan found zero XB_ building groups intersecting road triangles within 120 units of this spawn, with no collision flags during the short drive. This does not validate the rest of the city or retail physics. The gray body paint and dark glass remain unfinished.
A fresh host Chrome 150 WebGL2 run reached City scene ready with 43,607 nearby triangles, all 169/169 city textures uploaded, 832 routes, 12 ambient vehicle models and seven rendered traffic cars. The road diagnostic reports zero generated markings: the visible yellow and white lines come from authored road materials and textures. A 1.8-second W hold moved the 240SX about 5.44 m to 11 mph in first gear at 1,689 RPM. Chrome captured no page exceptions or console errors. The frame shows the road and ambient traffic in the correct orientation. The body is still flat gray and the windows dark; this is one L4RA spawn, not proof of citywide placement or retail handling.
Open full-size capture · Read runtime measurements and limits · Open this L4RA view
The XP guest GPU is not used for this check. Host Chrome reached City scene ready with 43,607 nearby triangles, all 169 textures, 832 routes, and six rendered traffic cars. The probe now captures city batches with the same local fixture and headlight inputs as the main city pass, after drawing its backdrop unlit. The normal driving screenshot shows no clear body-paint improvement; it remains pale gray with dark glass. Forcing the probe onto the body verifies that the local cube is populated, but produces a dark diagnostic rather than a plausible paint calibration. This fixes a lighting mismatch in the reflection capture; it does not complete the car appearance.
Open normal drive capture · Open reflection diagnostic · Read the implementation and limitations · Open this L4RA view
The XP guest GPU is not used for rendering. A full static scan found 34,757 of 37,233 traffic-route samples supported by exported L4RA pavement. The preview now requires road support for the selected traffic spawn and at 5 m intervals across the next 30 m; unsupported candidates are rejected. The packaged host Chrome capture reached City scene ready with 223/223 nearby textures, seven traffic assignments, and no browser errors. This filters obvious unsupported route spans but still uses static route samples rather than the retail traffic controller. The 240SX paint and glass remain unfinished.
Open full-size capture · Read the route scan and limits · Open this L4RA view
Host Chrome checked 86,285 road-supported samples across all 832 L4RA traffic routes against the region's facade collision geometry. The normal 1.9 m player-center check flags one Harbor neon/wall group along overlapping route samples; its capture shows a close roadside wall but no facade crossing the lane. A wider ±1.2 m footprint check adds a water-treatment structure at route 1020's diagnostic start; the screenshot shows a cramped industrial viewpoint, not the default spawn. These are candidates for closer inspection, not proof that source placements are wrong. The default start remains route 450. No instances or route data were moved.
Open Harbor capture · Open water-treatment capture · Read scope, measurements, and limits · Open the default L4RA drive
The driving view now draws nearby authored free-roam route samples on a small circular map in the lower left. The map rotates with the car and keeps its marker centered; it uses the selected region's traffic route graph and does not mix in event race paths. A fresh packaged Chrome 150 WebGL2 run loaded all 223 nearby textures, completed a short W drive, and reported no page or console errors. This adds local orientation for driving, not a full retail navigation map or connected-city map.
Open full-size drive capture · Open this L4RA drive
The player-up map now snaps a click to the nearest nearby route sample only when the authored road-surface index supports it. In this capture a click snapped within 1.3 m to L4RA route 740; after a short W hold the blue destination and dotted straight-line guide remained visible. Right-click clears it. This is a local waypoint aid: it does not steer the car or claim to implement the game's route planner. The packaged scene reached City scene ready with all 223 textures and no browser errors.
Open full-size capture · Open this L4RA drive
Chrome/WebGL2 loaded the packaged L4RA viewer and all nearby city textures without page errors or failed requests. A four-second W hold moved the player from its start position to 17 mph while six ambient traffic actors were rendered. The view confirms pavement markings and supported driving, while also making remaining defects visible: the car paint is still a flat neutral fallback, several traffic bodies look like unshaded white blocks at this distance, and the far junction has an unfinished dark gap. This is a working reconstruction, not a completed or retail-equivalent export. The XP guest was not used for rendering.
Open full-size capture · Open this L4RA drive · Read why the XP image is used for static analysis only
The packaged viewer completed its loading screen in host Chrome with no page exceptions or failed requests. A four-second W hold moved the 240SX to 21 mph; the HUD showed 3,981 RPM, first gear, and active engine audio. The XP guest was not used for rendering. This capture confirms that the build is running and drivable, while showing that the 240SX body still has flat gray paint and dark glass, and the city still needs further visual work.
Open full-size capture · Open this L4RA drive
The packaged page reached City scene ready, loaded the selected 240SX and 12 ambient body models, and accepted a four-second W hold to 13 mph. Chrome reported no page exceptions, console errors, or failed requests. The capture shows the road markings and drivable corridor; it also confirms the player paint is still a flat fallback and distant traffic bodies still lack their runtime paint state. The XP guest was used only as an asset/executable source, not for rendering.
Open full-size capture · Open this L4RA drive · Read why stock paint state is still unresolved
The loading cover remains until all 223 nearby textures and scene setup complete. After it cleared, the controls responded, the selector contained all 29 stock player cars, and the selected 350Z moved to 22 mph in the capture (25 mph at the end of the throttle hold). Chrome requested the packaged L4RA chunk manifest from the correct package path and reported no page exceptions, console errors, or failed requests. The screenshot still shows unresolved flat paint on the 350Z and pale fallback paint on distant traffic.
Open full-size capture · Open the same 350Z drive
?carShader=retail-fallback · same viewer and cameraBoth captures use the packaged L4RA scene in host Chrome after the loading cover cleared. The installed DEFAULT state contains effect parameters, not a direct body color; the 240SX paint texture slot has no matching bitmap in the installed car catalogs. The opt-in shader path uses the executable's decompiled car equations with a conditional fallback harmonic profile, but its identity transforms are unverified and it makes the body nearly charcoal. I left that experiment opt-in and did not replace the default paint. A live selected paint/material binding is still missing.