Univerce: making a living Unreal project reviewable

Gamibase built a project-scoped view of modules, targets, plugins, symbols, classes, assets, routes, build configuration, and test surfaces. The point was not to create another inventory document. The point was to give every later task a shared starting context.

In one measured internal snapshot, project intelligence indexed 108 modules, 1,949 files, 932 classes, and 2,589 functions in 4.1 seconds. Those figures describe that specific repository snapshot, including plugin modules; they are not a permanent product benchmark. Their value was operational: a review worker could begin with the project shape already known instead of spending most of its time rediscovering folders and build files.

The Univerce features examined in these campaigns often crossed several systems. Gamibase was used to reduce those goals to an explicit owner set, non-goals, expected artifacts, tests, and review gates before implementation began.

An artboard-media slice is a good example. The work was bounded to image import, local and network media playback, a single control surface, and deterministic URL classification. Codec development and multiplayer media synchronization were explicitly out of scope. The implementation plan separated URL resolution from rendering and decoding so classification could be tested headlessly. Independent review then caught integration gaps around media-module enablement, actor overlap, audio behavior, invalid path handling, and test coverage before sign-off.

Gamibase did not replace the engineering decision. It made the decision inspectable.

Maps and Blueprints are not ordinary text files. Univerce became a proving ground for the Gamibase UE Bridge workflow: inspect, preview, confirm, apply through the engine-facing mutation contract, export again, and compare semantic state.

In one scene-maintenance campaign, a batch of map and Blueprint changes was applied twice to test idempotency. The second pass had to preserve the intended values without duplicating actors, repeating transform offsets, or compounding changes on the targeted set. Snapshot evidence showed that exactly one map and one Blueprint changed while 4,116 other recorded project entries remained unchanged. The run also verified that no direct filesystem write was made to the encoded assets; UE Bridge performed the map and Blueprint mutations through the engine-owned path. Whole-map equivalence was not claimed because unrelated point-of-interest spawning produced known map-level drift outside the targeted actor set.

The important result was not the particular light or transform adjustment. It was proof that a bounded engine change could be planned, reviewed, applied, re-applied safely, and read back without treating a `.umap` or `.uasset` like an opaque file.

Gamibase also helped distinguish compilation success from release readiness. A recorded Univerce pipeline run proved both Development and Shipping builds without requiring source patches. Cooking did not clear its execute gate: the recorded attempts contained configuration and reference warnings, timeouts, and persistent process contention, but not enough terminal evidence to declare one closed root cause. Packaging and smoke validation were not attempted, and a stale executable was explicitly rejected as proof.

That is a valuable outcome. The workflow found the boundary between “the code builds” and “the product is package-ready” and refused to erase it with a green summary.

Completed tasks fed facts, decisions, warnings, and evidence references back into project-scoped memory. Later sessions could receive fresh project structure, class hierarchy, asset inventory, route inventory, and learned conventions in their briefing.

In a measured warm-start audit, combined orchestrator and worker activity entries fell from 309 to 118, while orchestrator activity entries fell from 232 to 64. That internal comparison represents 62% fewer entries overall and 72% fewer orchestrator entries for the measured worker-heavy audit. It does not imply the same gain for every task. It demonstrates the narrower point that reusable project context prevented substantial rediscovery in that workflow.