Two products, one night
Both vehicles opened the session unfinished. The M6 sat at a mechanical checkpoint from three weeks earlier, needing two narrow fixes: idle audio and high-speed steering. The Smart-Fifth sat at its mechanical-prototype lock, coupling proven, cosmetics untouched, and the owner’s verdict on seeing it again was one word: “bland.” Neither ask was “release these.” Both ended the night released and locked — WC M6 Gran Coupe v1.1.0 and WC Roadmaster Smart-Fifth v1.0.0 — and it is worth writing down honestly how that happened, including the parts that went wrong.
What went right
The pattern that held across both vehicles was small passes, each one judged live before the next began. The Smart-Fifth went through five approved passes in one evening — surfacing, rolling stock, refinements, lighting and exhaust, a reverse-rule fix — because each pass was small enough to actually look at in game rather than trust on paper. The same discipline that makes the coupling DLL and the steering DLL check the exact game version and refuse to guess when it doesn’t match paid for itself directly: it is what turned a silent two-week-old regression into a five-minute fix once someone finally read the log line it had been printing the whole time.
Reference discipline mattered too. The Roadmaster’s wheels went from looking like lawnmower blades to a real forged ten-hole rim because the geometry was rebuilt against zoomed reference photographs instead of memory of “what a truck wheel looks like.” And nearly every real defect this session was only visible at night, from the driver’s seat, in the actual game — not in a proof render, not in a build script’s success message.
What went wrong
Told honestly, in the order it was found:
- The Smart-Fifth’s coupling was dead, and it had nothing to do with the day’s surfacing work. The game had moved to a new build two weeks earlier; the coupling code’s own version check had been failing closed since then, printing a warning nobody had read. The trailer also shipped doubled tire and landing-gear geometry from a filter that didn’t get reapplied after a rebuild — which is what made it look like it was leaning.
- The tractor’s markers, brake lamps and turn signals had never worked, since the very first cosmetic version. Not a regression from this session — a defect that shipped for weeks. The lens colours were right; the one keyword that lets a material glow at all was simply never compiled in. No render catches that. Only a dark scene does.
- A disabled engine setting silently ate the exhaust. The project’s particle system had been switched off somewhere along the way, so the first attempt at smoke stripped itself out at build time while the build tool reported success. Caught only by opening the finished result with a second, independent tool and counting what should have been there.
- The M6’s tail lamps had the identical defect, and fixing it took three real attempts, not one. The first lit only half the lamp — the trunk lid carries its own copy of the lamp geometry under different names, so it fell outside the first fix entirely and the car showed one lit lamp and one dead one side by side. The second fix unified them and added reverse and brake state, but introduced a new bug: a light meant to sit behind the lamp was positioned using the wrong local frame of reference and ended up pointed into the cabin — the owner’s own screenshot showed the back seat glowing red. The third attempt fixed it by using the vehicle’s own, already-correct orientation instead — the same technique already sitting a few lines away in the same file, applied inconsistently the first time. Worth naming as a mistake, not just a fix.
What we’d do differently
A night test should be a standing step for any lamp work, not a lucky accident — every defect that survived to a “finished” build did so past both daylight renders and a green build. Checking the actual compiled result with an independent tool, not just trusting the tool that built it, should run on every package before it ships, not only after something looks wrong. And when one fix uses a technique that works, the rest of that same patch should be checked for the same mistake before it goes out, not after the next screenshot finds it.
Why it actually finished
Two things made the difference between starting this and finishing it. Working directly against Blender, Unity and the game, rather than through a slower, more formal command pipeline, is what made it possible to rebuild, reinstall and boot-test a fix in minutes — which is the only reason three real attempts at the M6’s lamps could happen in one evening instead of one. And nothing shipped on a claim: every install had a backup first and a live boot-check after, every release recorded the exact file hashes that went out, and every defect in this report is checkable against a specific log line or a specific screenshot rather than a memory of how the night went. That discipline is the actual reason two unfinished vehicles left the same evening released, locked, and correct.
Both are on the Downloads page. What they still owe, on request only: release-quality screenshots for the listings, LOD and damage-state meshes for both, and the M6’s lamps rendered against the real car’s own reference photographs rather than the source model’s stock lens shapes.