Execution resource model (normative)¶

Date: 2026-05-25
Companion: Native execution surface
Plan: .cursor/plans/compiler_lint_and_parallelism_330a4d8e.plan.md — Phase G

Four axes (do not conflate)¶

Axis	User surface	CLI / config	Must not use
SIMD / vector lanes	`@vectorized(lanes=N)`, `simd[T,N]`	reject `--vector-lanes`	`--cores`, `--jobs`
Runtime parallelism	`parallel for`, `@parallel(disjoint=…)`	`--cores`, `--threads-per-core`; `[execution]`	`--jobs`
Compile / check farm	not in `.li` source	`--jobs`, `--max-memory`	`--cores`, `@vectorized`
RAM budget	optional embed	`--memory-budget-mb`	thread team size

[execution]
cores = 8
threads_per_core = 1
memory_mb = 4096

Flags win over manifest; env vars deprecated.

MD across cores     → parallel for + --cores
dot in one thread   → @vectorized(lanes=8)
fast li-tests       → run_all.sh -j + --max-memory