Compiler roadmap — optimize execution without returning semantic power to Rust

[juv4uk]

Owner vision

my-lisp has only just taken semantic authority away from Rust; future performance work must never trade that authority back for speed.

The compiler is therefore an execution/optimization servant of Lisp semantics, not a new language definition.

Roadmap

• Compiler authority — make semantic ownership non-negotiable #66 — compiler authority boundary: make semantic ownership executable and fail-closed.
• Compiler oracle corpus — freeze observable semantics before lowering #67 — compiler oracle corpus: freeze the observations every compiled path must preserve.
• Compiler IR v0 — provenance-bearing semantics-neutral lowering data #68 — IR v0: provenance-bearing, semantics-neutral lowering data.
• Compiler first witness — execute compiled code with exact meta/native parity #69 — first compiled witness: exact meta/native/compiled parity on a nontrivial slice.
• Compiler optimizer v0 — semantics-preserving transformations with witnesses #70 — optimizer v0: admit one transformation only with explicit safety conditions and witnesses.
• Compiler deep gate — nightly meta/native/compiled semantic parity #71 — deep compiler gate: fast PR parity + nightly expensive falsification.
• Compiler substrate parity — one semantics across native, WASM and future FPGA #72 — substrate parity: one semantics across native, WASM and future FPGA observers.

Non-negotiable invariant

Lisp owns meaning
      ↓
semantics-preserving transformations
      ↓
compiler / optimizer
      ↓
Rust | native | WASM | FPGA
      ↓
same contractual observation

Performance improvements are welcome only when the semantic evidence remains at least as strong as before.

Success criterion

We can make my-lisp substantially faster over time while still being able to point to executable evidence that no optimization/backend has become an undeclared semantic authority.