Debugging, validation, and profiling

Graph-debug procedure

1. Enable node Debug on the earliest suspicious stage.
2. Inspect the Data View and Attributes List: count, bounds, density, transforms, seed, attributes.
3. Move the debug point downstream one stage at a time.
4. Disable spawners while validating data.
5. Use graph/node profiling to identify CPU cost; use Unreal Insights for system-level cost.
6. Cleanup and regenerate to eliminate stale output.

Runtime tools

• pcg.RuntimeGeneration.EnableDebugOverlay 1: scheduler/runtime overview.
• pcg.GraphExecution.DebugDrawGeneratedCells 1: sources, generating cells, grid/coordinates.
• pcg.RuntimeGeneration.EnableDebugging: verbose scheduler logging.
• pcg.RuntimeGeneration.Refresh: cleanup and schedule regeneration.
• pcg.RuntimeGeneration.Enable: freeze/enable runtime generation.
• pcg.GPU.FuzzMemory: expose uninitialized GPU output assumptions.

Debug overlays are not generally available in Shipping and may differ by build configuration.

Validation matrix

• empty input and zero points;
• smallest and largest bounds;
• steep/vertical/overhanging surfaces;
• overlapping exclusion volumes and Data Layers;
• seed zero, repeated seed, and changed seed;
• missing asset/table row/soft reference;
• cleanup -> regenerate -> reload -> cook;
• source control clean checkout and commandlet generation;
• target platform scalability/feature-level switches;
• fast traversal/teleport and multiple runtime sources;
• maximum shipping density plus HLOD/nav/collision build where required.

Common failures

Symptom	First check
Points exist but nothing spawns	mesh/actor selector attribute, bounds, node enabled, output path
Same random choice repeats	duplicated seeds; place Mutate Seed at intentional branch
Content leaks into roads/buildings	exclusion input/bounds and Difference mode/order
Graph regenerates constantly	tracked actor/component changes or circular dependencies
Runtime pop-in	generation radius, bounds modifier, scheduler budget, traversal speed
Cleanup leaves stale actors	generation ownership, partition actors, grid change cleanup sequence
GPU path slower	transfers, small data, compute-graph fragmentation, preparation cost
Packaged result differs	editor-only input/node, unsupported platform path, unbuilt persistent output

Performance record

Record point count, spawned instance/Actor count, CPU node times, GPU execution/memory, generation latency percentile, cache memory, cleanup time, streaming overlap, and final render cost. PCG graph time alone does not include the lifetime cost of what it spawned.