Research-Stack/2-Search-Space/FAMM/docs/FAMM_MODULE_STATUS.md
2026-05-11 22:18:31 -05:00

2.5 KiB

FAMM Module Status

Status: HOLD_DOCUMENTED_NOT_DEAD

Claim boundary: this document resolves the dead-code audit ambiguity for the FAMM Lean surfaces. It does not promote every FAMM file to an active build target. It records which files are canonical, which files are fixtures or experiments, and which proof debts must close before promotion.

Current Files

File Status Decision
FAMM.lean active search-space surface keep as current canonical FAMM module in this directory
FAMM_minimal.lean minimal lemma/proof fixture HOLD; keep until its lemma fixtures are either ported or explicitly retired
FAMM_refactored.lean refactor/test-vector candidate HOLD; keep with FAMM_refactored_test_vectors.json until owner selects canonical replacement
FAMM_hyperheuristic.lean WIP extension HOLD; standalone typecheck passes with no sorry, but do not promote until its stronger performance-bound claims are receipted

Why Not Delete

The audit correctly found that FAMM_minimal.lean and FAMM_refactored.lean are not imported by the active search-space module. That is not enough evidence to delete them, because this directory also contains proof-space mapping, test-vector, and Verilog fixture material.

The permanent decision is:

unimported != dead
unreceipted != active

So the files are retained as documented HOLD surfaces until a later pass either:

  1. ports their unique fixtures into the canonical module,
  2. moves them under an explicit archive/fixture path, or
  3. removes them with owner approval and a receipt.

Proof Debt

FAMM_hyperheuristic.lean now typechecks without sorry with:

cd 0-Core-Formalism/lean/Semantics
lake env lean ../../../2-Search-Space/FAMM/FAMM_hyperheuristic.lean

The closed local laws are:

  • out-of-bounds hyper-heuristic adjustment fails closed,
  • switch count is monotone,
  • a concrete LUT entry replays to its projected best heuristic.

The remaining documented proof debt is not a sorry in source. It is the stronger performance-bound claim:

LUT lookup existence does not by itself prove a frustration bound against
missPath.performanceHistory.

Those are real theorem-shape issues, not syntax cleanup. The next Lean pass should split over-broad theorem statements into smaller replayable laws before trying to close them.

Receipt

Machine-readable status receipt:

shared-data/data/stack_solidification/famm_module_status_receipt.json