The engine stands fully on its own + richer materials

The watershed of standing alone: from "borrowed shell" to "standing on its own"

This is a watershed day in the engine's history. All along, the engine had kept a low-level dependency on a mature external general-purpose rendering library — using it as the "reference answer" to check whether its own rendering was correct. This day removed that last dependency entirely, and the engine now runs solely on its own rendering pipeline. This isn't an ordinary dependency removal but a shift in posture: from "standing on others' shoulders, always referencing someone else" to "being the standard itself." The items below — a self-check benchmark, richer materials, device compatibility, a native preview — are all the accompaniment around this "standing alone," together declaring that this engine has grown into a mature renderer that stands on its own.

A built-in pixel-parity benchmark: verifying itself

After removing the external library, a practical problem followed: previously that library was the reference to confirm "is the rendering correct," and without it, what guarantees quality? The answer: the engine carries its own "pixel-parity" benchmark — it generates its own reference answer and checks each rendered frame pixel by pixel, alarming on any deviation. That turns the "quality judge" from externally borrowed into built into the engine itself. Its far-reaching significance: from now, the engine's image quality, performance and feature evolution are fully in its own hands, no longer led by an external library's capability boundaries and trade-offs — whatever you want to add or optimize, no working around someone else's limits. An engine that can verify itself is truly qualified to evolve independently long-term.

Tier-1 PBR materials: emissive + ambient occlusion + tone-mapping looks

The material system was unified and stepped up a level this day. The low level merged what had been several material variants into one unified "render-pass"-based material model, simplifying concepts and making room to add new capabilities. On this unified model, several realism-boosting abilities were added: emissive, letting objects glow on their own for neon signs, glowing runes and energy cores; ambient occlusion, naturally darkening crevices, corners and contact points to greatly strengthen depth; and several tone-mapping looks to choose the overall feel of the image (cool, warm, cinematic, etc.). To keep "more material variety" from "costing frame rate," these complex materials are threaded through a render-pipeline cache, reusing the same and keeping different ones separate. All told, things in a scene look more "right" and more tactile — the key upgrade from "can render" to "renders beautifully."

Wider device support: the device layer completed + fallback without advanced features

This day also pushed "letting the engine run on more machines" forward. The low-level "rendering device layer" completed a key full set of capabilities, so the engine can fully request and use the device resources rendering needs in the browser, no longer missing pieces; on the fallback path without the latest GPU features, the remaining gaps to "renders properly" were wired up one by one. Even more thoughtfully, for devices missing certain modern GPU features, a material-compatibility route was built so they can draw objects correctly even on a simplified path. Net result: the range of machines that can run 3D properly widened another notch. For a mass-market platform, "runs on weaker machines" matters as much as "looks better" — the former decides how many can get in, the latter whether they stay.

A native preview player — a game in fifty lines

Preview games this day moved to the platform's own native runtime host: a clean game entry and loader responsible for loading and running a game correctly, with no detour through an external host. The game template was compressed to about fifty lines — meaning writing a runnable game starts from fifty lines. The value here is "a light enough starting point": a beginner or an AI faces not a big lump of framework code to understand first, but a minimal, clear skeleton ready to build on immediately. The lighter and more standard the start, the smoother the road of "chatting a game into being" — AI generating code from this clean start makes fewer errors, and a human taking over finds it easier to read and edit. A native host plus a minimal template is a step toward cutting the entry bar of "making a game" to the lowest.

Stability: HTTPS everywhere + preview auto-recovery

This day also added two stability touches. One is support for a "HTTPS everywhere" dev mode — some modern browser capabilities (including certain graphics and device features) are only granted under a secure origin, and HTTPS makes them properly available while better matching a real deployment environment. The other is preview auto-recovery: when the preview errors due to a transient backend hiccup, it tries to reconnect and self-heal on its own, rather than sitting on an error page waiting for a manual refresh. Both are the work of "not interrupting you" — what most breaks flow during development is often not some big bug but these small recurring issues needing a manual rescue every so often. Automating them away is what makes sitting down to make a game for hours truly smooth.

What this day means

The core of this day is the engine completing its "independence." Removing the last external rendering dependency and carrying its own pixel-parity benchmark means it grew from "running on a borrowed shell, referencing others" into a mature renderer that stands on its own and verifies itself. On that footing, materials rose to real tactile quality, device compatibility reached weaker machines, the preview got a minimal native starting point, and stability became more resilient — each an extension that means more after this independence. A platform daring to make its most core rendering foundation fully in-house, depending on no external library, is after holding its own fate: from now, how far image quality can go, how much performance can be optimized, and what new features to support are no longer at others' mercy. That's the weight this day most deserves to be remembered for.