Build a phase-level benchmark, remove redundant recording work, and locate the command-pool reset cost that a future recording thread could own.

Prove that fireEngine can record geometry into secondary command buffers under dynamic rendering before treating that structure as a performance opportunity.

A retrospective map of the questions, controls, measurements, and limits that turned parallel command recording into a conditional policy in fireEngine 0.9.

Keep fireEngine's public renderer facade small, organise Vulkan owners by lifetime, and verify the complete imported-scene path through bounded device scenarios.

Replace fireEngine's swapchain-compatible resources as one ownership group, retire presentation explicitly, and preserve compiled scene resources.

Advance imported rotation channels into current scene transforms while keeping render preparation and compiled GPU resources unchanged.

Turn fireEngine's textured scene into a coherent opaque 3D path with a Vulkan projection, an extent-matched depth attachment, and correct culling.

Compile selected RGBA8 descriptions into device-local Vulkan images and samplers, upload them safely, and bind one base-colour texture per draw.

Compose fireEngine's Vulkan-free scene data and import a deliberately narrow glTF slice without leaking the source format into the engine model.

Add Vulkan-free image, texture, material, animation, and scene-component descriptions while preserving fireEngine's explicit preparation boundary.