FORMALIZATION_COVERAGE.md

Model-Theory Normalization

FORMALIZATION_COVERAGE.md

Download original file

Original Markdown with line links

Formalization coverage — v0.10

PASS: 9 new local Lean modules, 105 theorem declarations (78 generic and 27 concrete kernel examples), 50 definitions, 2 abbreviations, 1 structure and 1 inductive type: 159 explicit named declarations audited. Generated constructors/fields compile but are not counted again as named research declarations. 17 expected-outcome tasks passed: 9 cold builds, one full axiom audit, one positive control and six rejections. Only propext, Classical.choice, and Quot.sound appear. No custom project axiom, sorryAx or Lean.ofReduceBool is accepted. Library warnings are errors.

Module Exact checked scope
BinaryWords Word semantics, padding, full-adder/subtractor truth tables, ripple operations and length/value theorems
BinaryMulDiv Actual shift-and-add multiplication and restoring long division, positive-divisor invariants and cost recurrences
BinarySigned Sign/magnitude arithmetic, zero detection, Euclidean adjustment and safe divisor rejection
BinaryEncoding Integer/word boundary conversion correctness and semantic scalar adapters; conversion costs excluded
BinaryCosted Actual metered evaluators, erasure theorems, exact counts, modular multiplication and quadratic width budgets
BinaryPricing Explicit unsigned request interpreter, trace semantics and conditional length/width cost bound
BinarySignedCost Metered sign handling, zero scans, signed addition/multiplication/division and their cost bounds
BinaryBackend Signed subtraction, modular addition, final scalar semantics and budgets
BinaryExamples 27 finite direct kernel computations, including negative division, padding, carry, borrow and widened modular products

The current Lean modules import pinned Mathlib directly. The predecessor archive is preserved but its historical local Lean library is NOT recompiled by this block. The selected Python baseline payloads are exact original bytes; their 203 source profiles are re-executed independently and under the new port.

Boundaries

No universal theorem lowers the full discovery loop into these binary routines. The 203 complete bit-backed discovery executions are PYTHON tests, not 203 kernel-checked whole-loop instances. Their scalar algorithms have generic Lean counterparts; their parser/port/transcription is not verified extraction.

The cost theorem concerns a declared gate/cell model. It excludes input conversion, allocation/indexing/queue work, compiled runtime, proof checking and bit-machine implementation overhead. Primitive arithmetic correctness and priced scalar recurrences do not alone close that missing implementation refinement.

The conditional unsigned trace theorem requires an explicit trace and explicit operand-width/length bounds. The signed primitive theorems are separate; no unproved universal trace-composition premise is silently discharged.

Lean 4.34.0-rc2, commit 6a10ac8c22beadecabdbb0919c2b50214762f91d. Mathlib aeb72f1e9981b6d35b426b16dca7759be3e7ba06. Existing upstream cache reused, not rebuilt. Interpretation and implementation review are self-review.

Original file and source record

This copy matches the recorded source file. Its publication here does not change the work’s review status.

SHA-256: 7f8e084736e493aa906eb71983dac4bccaa143ca334c6740a25517fe163fb308

Original source on GitHub

Back to Model-Theory Normalization