Decoder check pack

Claim
On the EPFL benchmark "dec" (8-input decoder), a 6-input LUT network of 258 LUTs at 2 levels, against the
published 264, plus a certified proof that 257 LUTs do not suffice (SAT refutation checked by cake_lpr, plus a
Lean 4 proof). Also exactly 129 LUTs for the 7-input decoder.

Replay, one command
    sh replay.sh

Contents
    netlist/dec_258.blif         the 258-LUT network (258 LUTs, 2 levels)
    simulate_dec.py              exhaustive simulation against the benchmark and structure check
    bound/bound_257.cnf.gz       the formula "257 LUTs suffice"
    bound/bound_257.lrat.gz      its LRAT refutation
    lean/                        Lean 4 development: lowerBound (258 <= nodes for 8 inputs),
                                 lowerBound7 (129 <= nodes for 7 inputs), the Lean-written formula
                                 bound.cnf.gz and its refutation bound.lrat.gz
    replay.sh                    the replay
    SHA256SUMS                   SHA-256 of every file in this pack

Hashes of the decompressed formulas and refutations
    bound/bound_257.cnf    cc7fb8f24aa9cec2ed7bb1524c28dc48b33eaa010890a0f2084e65e92b4c636d
    bound/bound_257.lrat   c885e37d8cebc29c46a268555b766e0cc47cbc447eb150e5111155acde0950d7
    lean/bound.cnf         6625d4f4cf15251a6d51de4c4f91fd1b4b5e86c61d2f0fad4f50ec7fe5f0b727
    lean/bound.lrat        95fc046950c012b073edba638ea04fcdad3206a5eaed51c8fda9154dcc3b2c10
bound_257.cnf has CRLF line endings; the hash above is of the file as distributed.

Checker and tool versions of the recorded runs
    cake_lpr on bound_257      binary sha256 98c1649dc01f6ba38e4424beeff1b4681069b6175d90047b8c2c73316cd9ef2f
                               answer "s VERIFIED UNSAT"
    cake_lpr on lean/bound     commit a36874a8b750b43fe4b385b8ddbf5b033e46a3fa,
                               binary sha256 72cd6fc6d93b5be1fd7de746113b033fa88c54e80c46a3d28f4aa6a3ec179c5d
                               answer "s VERIFIED UNSAT", 4.0 s
    CaDiCaL (bound_257)        binary sha256 4c9e2a21d9eab0a3e72a411b9e732f30655238d72220b1952a3b3bb0602d30ff
    CaDiCaL (lean/bound)       rel-3.0.1, commit c60730422e758ef1cebe7aeddf2dda31c996bf04, answer UNSAT in 8.5 s
    Lean                       leanprover/lean4:v4.35.0
    Mathlib                    tag v4.35.0, commit 56a005855ea17e0b0958f73ec801ae140b0a1679

What the Lean build ends with
    'DecoderOptimality.lowerBound' depends on axioms: [propext, Classical.choice, Quot.sound,
     DecoderOptimality.Certificate.cert_ok._native.native_decide.ax_1_1]
    'DecoderOptimality.lowerBound_general' depends on axioms: [propext, Classical.choice, Quot.sound]
    'DecoderOptimality.lowerBound7' depends on axioms: [propext, Classical.choice, Quot.sound]
The fourth axiom of lowerBound records that Lean's compiled evaluation of its built-in LRAT checker accepted the
embedded refutation (native_decide). There is no sorry.

Trust boundary
cake_lpr verifies that each LRAT file refutes the formula it is given. The Lean proof covers every step from the
definition of a 6-input LUT network computing the 8-input decoder (lean/DecoderOptimality/Model.lean), through the
lemmas and the Lean encoder, to the refutation of lean/bound.cnf. Outside the Lean proof: that Model.lean states what
you mean by a LUT network (read it), Lean's kernel and, for the one native_decide evaluation, Lean's compiler; the
encoder that wrote bound/bound_257.cnf (its refutation is checked by cake_lpr, and its correctness rests on the
written proof); and the 258-LUT netlist, which is checked against the benchmark by exhaustive simulation
(simulate_dec.py), outside Lean. The 129-LUT upper bound for the 7-input decoder is one shared LUT over inputs 0 to 5
with table 0x000ff0ffff0ff000 plus one LUT per output; lowerBound7 is the matching lower bound.
