The fabric is the primitive for distributing intensive work: fan a command or a kernel over N inputs across isolated instances, all at once.

Fabric.map("resize", frames, width: 16)          # 16 concurrent isolated workers
Fabric.map_kernel(decode_wasm, frames, width: 8) # a pool of persistent kernels

Two knobs, one surface: width (how many run at once — throughput) and tier (how isolated each is — containment). They're the same primitive seen two ways: distribution is "many contexts," isolation is "how contained each context is."

This is what a CPU renderer is on this platform — not a monolith, but a graph of kernel toolkits with frames fanned across them. And at the node tier the two knobs merge: a worker can be a peer VM on another machine, so fanning out and isolating are the same move.

• The depth knob: Isolation tiers & trust

• The hot-loop shape: The six shapes