Skip to content

Evidence, reproducibility, and open obligations

An auditable register of source mathematics, GCL finite checks, conditional deductions, editorial admission, and work still required.

How to interpret the evidence grades

Grade Meaning
EXT Established mathematical theorem imported with attribution
CAL Finite calculation or elementary argument, with a written proof or executable check
SRC Source transcription, theorem-interface audit, or parameter ledger
COND A conditional deduction from a published theorem under explicitly stated hypotheses
OPEN No proof or complete implementation currently admitted

A document merge or public MkDocs page certifies neither mathematical truth nor independence. The independent specialist certification route remains separate.

Source-semantic update, 1 November 2026: WP01 source erratum and classical transfer results distinguishes the printed statement of Theorem 12.1 from its final displayed proof chain. The published printed theorem has a stronger second branch; the record uses the proof-backed bound pending reconciliation. Exact-code regression checks and a restricted explicit-table SAT transfer are recorded in MATHSOLVE PR #1152. Neither a full quantum compression implementation nor an unrestricted SAT implication has been established.

WP03: executable parameter-recurrence slice (1 November 2026)

The admitted WP03 source and execution receipt document a standalone Lean 4 numerical implementation of the recurrence used for the compression pipeline's size parameter. The exact Lean source, SHA-256 540f7c27755c2b74c3b5f2b5001134fca65c634962dc9cb82ea93b5ff651b380, was read back byte-for-byte from the research branch and executed successfully (exit code 0). At input n=3, its first five values are 7, 64, 16,384, 1,073,741,824, and 4,611,686,018,427,387,904. Lean guards also checked 126 finite inequalities over 0 ≤ c ≤ 5 and 0 ≤ n ≤ 20.

The upstream formalization is pinned to commit 5926bdff and requires Lean 4.35.0-rc3 plus its locked dependencies. The standalone replay used Lean 4.34.1 and does not import upstream MIPRE, prove its PolyTimeFun complexity certificate, execute DecisionCompiler.compute, or implement quantum introspection, answer reduction or repetition. Installing the upstream toolchain was attempted, but did not finish within the bounded retrieval window. This is an execution-environment/dependency obligation, not a mathematical or safety restriction.

Research status: precisely one verified numerical parameter subroutine, with no change to P-versus-NP claim status or to the open succinct classical transfer obligation.

WP04: native program semantics replay (1 November 2026)

The WP04 research PR adds standalone Lean source and an exact execution receipt for a source-aligned model of the seven MIPRE native program constructors. It copies the upstream Data, Prog, Env and timed Eval definitions into a separate namespace and executes an independently implemented, fuel-bounded interpreter on Lean 4.35.0-rc3. The ten finite guards and two term-level Eval witnesses passed. The two-step loop evaluated to nil with 12 native cost units; process exit code was zero.

The source is pinned to upstream commit 5926bdff. The executed file has Git blob 1e422f210df6db82538e36e5a99c6630aba6300b and SHA-256 f4419be14a6e5d913af5c47a41a5890b2191316fa4f9029474ed49bc9208d63e; GitHub readback matched the local executed blob. This is a finite CAL result and source-semantic rehearsal, not an imported upstream theorem, a general proof of interpreter correctness, an executed introspection compiler or a quantum compression result.

The required Lean version has now been installed. This dependency blocker was subsequently resolved in WP05. WP05 has since imported the actual MIPRE.Cost.Prog and built the upstream introspection compiler. Instantiating DecisionCompiler.compute on concrete positive parameters and testing a compressed verifier remain separate implementation tasks. No SAT transfer beyond the explicit-game result has been established.

WP05 update (2 November 2026): Research PR #1162 records successful builds of the upstream native model (1,219 jobs) and introspection-decider module (3,978 jobs). The imported Prog replay passed. Positive-parameter compression remains unexecuted.

WP06: imported compiler theorem replay (2 November 2026)

MATHSOLVE WP06 directly imports the pinned upstream MIPRE.Background.Introspection.DecisionCompiler and executes seven Lean theorem proofs under the exact Lean 4.35.0-rc3 environment. For parameter zero, the imported compute selects zeroProg; this result is well-scoped, has a source-Eval execution witness, and obeys the source's exact acceptance condition (empty questions and answer lengths at most two). Concrete empty and nonempty question cases were proved accepted and rejected respectively. A seventh theorem characterizes the parameter-one compiler branch symbolically, without executing a generated positive-parameter program.

The tested Lean source has Git blob 68443763c21f182f02f05a151addd38c14e576ff and SHA-256 d6c4242a5242157edae2426bd407ab46469fe0ad0c764865fd0362fde2783451. Its exact-branch Git blob matched the local executable file, and lake env lean exited 0 with no errors. The execution and scope receipt records the first corrected namespace error and clean replay.

Boundary: This verifies genuine source-interface theorems for the degenerate parameter-zero branch. No concrete positive-lambda quantum introspection output, three-stage compressor, classical asymptotic lower bound or SAT transfer theorem has thereby been established.

WP06 (2 November 2026). The admitted Lean proof replay confirms six zero-parameter source properties, with details in the WP06 evidence record. Corrected positive-parameter interface propositions remain unverified candidates, not an executed compressor.

WP06 source-audit update (2 November 2026): MATHSOLVE WP06 records the zero-parameter and oversized-description erasure behavior of the pinned introspection decider compiler. The resulting non-injectivity is a source-budget constraint, not a general SAT lower bound. Its companion Lean corollaries were subsequently replayed successfully against the imported source in WP07 (exit 0); the earlier description was accurate only at initial publication.

WP07 (2 November 2026): The source-erasure proof replay and adequacy theorem distinguish two cases. Descriptions whose encoded sizes exceed the selected budget can collapse to identical compiled decider programs. A positive budget at least max(esize S, esize D)+1 preserves both source descriptions through the compiler's clamp. Both are Lean-checked results against the pinned upstream compiler, but preservation of source descriptions is not compiler injectivity or a polynomial SAT transfer. The full quantum compressor and asymptotic classical lower bounds remain open.

WP08 (2 November 2026): MATHSOLVE native nine-bit decider and evidence implement the identity input game's decider component using actual imported MIPRE.Cost.Prog and MIPRE.Decider types. The exact Lean 4.35.0-rc3 file (Git blob 45b9fa687fad92bf1b1d4be2809473e0c4f9957d) compiles with a source-native binding/scoping proof, evaluates to an encoded program size of 145,459 nodes, and passes 512 diagonal nine-bit question checks, answer/length boundary tests, and malformed bit-tree rejection. One accepting example returns true in 63 source-model cost units. The fuel-based interpreter is independently implemented. Its soundness against the imported Eval relation was subsequently proved in WP09, but completeness, a formal TimeBoundAt bound, and a complete Sampler 9/Verifier 9 are not established. This remains a finite CAL result, not a compressed output, classical lower bound, or SAT transfer. See WP08's exact execution receipt.

WP09 (3 November 2026): MATHSOLVE PR #1179 adds a Lean-checked general soundness theorem for the same seven-constructor native Prog interpreter. For all fuel limits, environments, programs, results and source cost units, runFuel f e p = some (r,t) implies the imported MIPRE.Cost.Eval e p r t. Its exact source (Git blob 3b64353814fd70a77810e51f2d04cbcfb4271def) passed Lean 4.35.0-rc3 against the pinned upstream MIPRE library, with exit code zero and no diagnostics. The execution receipt records exact hashes, initial proof repairs and the local OS execution limits. This establishes valid source Eval witnesses for successful interpreter calls, not completeness of the interpreter, source Decider.Accepts equivalence, a nine-level sampler, positive-parameter quantum compression, or a SAT-to-classical strategy lower bound.

Source-to-claim matrix

Item Primary evidence Grade Maximum warranted claim
MIP* = RE Ji et al., v4 EXT Published complexity theorem; not P ≠ NP
Quantum low individual degree soundness Ji et al., arXiv:2009.12982 EXT Corrected source underpinning
Gap-preserving compression Theorem 12.1 EXT Conditional normal-form theorem for quantum strategy resources
Magic-square classical value \(8/9\) MATHSOLVE #1137 CAL Finite two-prover game bound
Source compression parameter ledger Protected ledger SRC Exact symbolic resource and error bookkeeping
Nine-level identity input verifier Protected note CAL Finite verifier semantics; not full compressor
512-pair exhaustive test Finite emulator CAL Reproducible finite local behavior
Double-exponential entanglement bound for compressed family Source theorem and WP00 note COND Requires encoded input satisfying the source theorem
Indexed language is cofinite and in P WP00 note COND Existence argument for a selected unary language, not arbitrary game-value computation
General SAT-to-classical strategy-cost transfer Research issue #1136 OPEN No such implication proved
P = NP or P ≠ NP Parent PNP-001 OPEN Neither proposition proved

Exact finite test

The protected Python emulator uses the standard library, represents the nine binary coordinates as tuples, and exhaustively checks all 512 vectors for both players and all nine marginal transitions.

It checks that each factor is the required coordinate basis vector, that each marginal is the prefix projection, that a level's linear increment reconstructs the next marginal, and that the diagonal pair distribution has 512 distinct support points. It also tests the specified empty-answer decider and representative invalid queries.

It does not: run the quantum compression construction, prove Theorem 12.1, produce a strategy at enormous Schmidt rank, or certify the source's Gödel-coded Turing machines.

Research genealogy

The bounded work began with PNP-ABOVE-WP00 and continued through finite magic-square calibration #1137, source interface #1140, resource semantics #1142, parameter ledger #1146, and nine-level input #1147. The last two reports have protected-main readback.

The source theorem version is arXiv:2001.04383v4, 16 October 2026, particularly Definitions 4.14 and 5.12, Theorems 8.3, 10.30, 11.4 and 12.1, Figure 15, and Table 2. A versioned URL is not a byte-level archive hash; do not claim a hash for the external PDF unless one is actually retained.

Falsification gates

A proposed transfer fails if it confuses Schmidt rank with classical time; relies on an expensive prover as though it were an ordinary SAT algorithm; replaces explicit input length with a succinct index without a reduction; relies on unprovided numerical constants; or implicitly assumes the SAT-to-strategy construction it must prove.

A claimed full compressed-game execution fails its evidence gate until there are exact source-faithful encoded machines, runtime and representation bounds, correct promises, effective constants, a functioning implementation of all three stages, and independent output checks.

Finite experiments may suggest lemmas but do not justify asymptotic lower bounds.

Open obligations and stopping conditions

  1. Source machine representation: Compile the nine-level input into the source's actual Turing-machine encoding, including malformed-input behavior and step bounds.
  2. Numerical constructivity: Obtain explicit sufficient values of \(\lambda\) and \(C_0\), or document why they cannot currently be extracted.
  3. Classical invariant: Define a nontrivial restricted model and one falsifiable quantitative lower-bound target.
  4. Uniform transfer: Supply the direction and complexity of a SAT-to-realization reduction, or demonstrate the exact obstruction.
  5. Barrier audit: Check relativization, natural-proofs, and algebrization assumptions before novelty promotion.
  6. Editorial continuity: Update the preprint record if the claims or source identities change. The parent programme's original formal-conjecture definition-level concordance issue remains outstanding.

A negative result with a precise counterexample and exact scope is a legitimate terminal for a bounded investigation. Neither documentation nor a closed work package implies the Millennium problem has been solved.

For an accessible explanation of these cautions, see the reader's primer. For formulas and proof interfaces, see the technical spine.