Start Here for Optimization Work

Applies to Vyre 0.7.2.

Read README.md before you change an optimizer or backend. It defines the two-layer boundary and the proof contract.

Implementation sequence

1. Find the owning boundary in OWNERSHIP.toml.
2. Identify the optimization class in TAXONOMY.md.
3. Check the generated pass reference in PASSES.md.
4. Keep semantic IR rewrites in vyre-foundation/src/optimizer/.
5. Keep target-specific lowering strategy in the owning driver crate.
6. Update OP_MATRIX.toml when operation or backend support changes.
7. Update BENCH_TARGETS.toml when a target or baseline class changes.
8. Prove semantic equivalence and exercise the real optimized path.
9. Run the required commands for the owning boundary.

Placement guide

Question	Owner
Does the rewrite preserve IR semantics for every backend?	vyre-foundation/src/optimizer/
Does it emit a target-specific instruction or API call?	The owning concrete driver crate
Is it neutral launch, binding, validation, cache, or residency policy?	vyre-driver/src/
Is it persistent queue, scheduler, or I/O behavior?	vyre-runtime/src/megakernel/
Is it benchmark measurement or reporting?	vyre-bench/ and BENCH_TARGETS.toml
Is it operation support or parity status?	OP_MATRIX.toml

Historical plans and reports do not override this control plane.