Services¶
Shared machines every task calls on — built once, never rebuilt per show.
Services are the always-on layer: each has an address, a health state, and a queue. (The distinction from a tool: a tool runs and exits — ffmpeg can't be "down," only missing. A service is up or down, and the facility notices.) Services generate the production record; Analytics answers questions from it.
Image conversion¶
One source, every context — EXR for finishing, PNG/JPG proxies, thumbnails for browsing — automatic.
Video conversion¶
One source, every context — ProRes for editorial, mp4 for web review, stills for boards — automatic. Each consumer gets its native currency. "Review media" is not a separate system; turntables, playblasts, and dailies are this service doing its job on the way to the review phase.
3D interchange¶
Ensure 3D compliance throughout all production tasks. One service owns 3D interchange — DCCs never negotiate formats with each other.
Production manifests¶
Data-rich manifests for every version, task, and asset — what went in, what came out.
Scene description (USD)¶
One scene description across every DCC.
Render / queue management¶
Farm submission and queue management — headless execution at scale: the same recipes, thousands of times.
Production tracking¶
Live status of every version, task, asset, shot.
Examples needed
Per-service: the endpoint shape, a worked request (source in → derived set out), and where each service is invoked from the task phases.