v11.0.0 DOI Concept DOI CI

This repository is one chamber of David Fox's Opera Numerorum: a growing collection of machine-checked arithmetic, geometry, and analysis in Lean 4. It approaches the Beal Conjecture through a tower of formal layers — from elementary divisibility, through elliptic-curve conductor arithmetic, to the modularity machinery pioneered by Wiles, Ribet, and Tate.

The aim is not to make a green build look like a finished theorem. It is to make every mathematical dependency visible, inspectable, and worthy of trust.

v11.0.0 — Conditionally complete

The final assembly Beal.Final.ConditionalBealTheorem exposes exactly five named mathematical premises. The finite matrix is ledger-derived; the absent geometric compatibility theorem is named rather than implied.

Andrew Beal (1997) observed and offered a prize for the following:

If $A^x + B^y = C^z$ where $A, B, C, x, y, z$ are positive integers and $x, y, z \geq 3$, then $A$, $B$, and $C$ share a common prime factor.

The condition $x = y = z = 2$ is excluded because Pythagorean triples (3ドル^2 + 4^2 = 5^2$) are coprime. The conjecture concerns the genuinely harder territory of higher exponents.

Beal's conjecture is strictly stronger than Fermat's Last Theorem. The implication is a clean descent argument.

Claim. Beal's conjecture $\Rightarrow$ Fermat's Last Theorem for $n \geq 3$.

Proof sketch. Let $n \geq 3$ and suppose $A^n + B^n = C^n$ with $A, B, C$ positive integers. This is a Beal triple with $x = y = z = n$. Beal's conjecture would then require $\gcd(A, B, C) > 1$: there exists a prime $p$ dividing all three. Write $A = pA'$, $B = pB'$, $C = pC'$. Then $$ p^n A'^n + p^n B'^n = p^n C'^n ;\implies; A'^n + B'^n = C'^n, $$ a strictly smaller solution. Infinite descent forces the conclusion: no minimal positive solution can exist, so no solution exists at all. $\square$

Wiles (1995) proved FLT unconditionally via a different route — the modularity of semistable elliptic curves. That Beal's conjecture would give an independent implication illustrates how closely the two problems are coupled at the level of conductor arithmetic and level-lowering. The formalization makes this coupling explicit rather than gestural.

The five final premises are:

1. J0DecompositionSoundness_26;
2. MwrankCertificateSoundness_26;
3. FormalImmersionSoundness_26;
4. FreyCurveExists;
5. LevelLowering_26.

The theorem does not prove these inputs. It proves that they suffice.

Level26_GeometryBridge.lean derives [ C_3=\begin{pmatrix}1&12円&1\end{pmatrix},\quad P=\begin{pmatrix}1&01円&1\end{pmatrix},\quad PC_3=M_3=\begin{pmatrix}1&10円&2\end{pmatrix}. ] Foundations v1.2.0 supplies a finite formal-coordinate calculation on a symmetric-square cusp chart. It does not construct the global Picard scheme or an actual Abel--Jacobi morphism, so QExpansionCotangentCompatibilityAtInfinity26 remains an explicit geometric premise.

The companion beal-level-26-foundations develops these missing foundations. Its closed formal cusp-chart differential and odd-prime valuation-input release is 10.5281/zenodo.22286222, under concept DOI 10.5281/zenodo.22272382. The corrected finite-only v1.0.1 archive remains 10.5281/zenodo.22272714. It is a companion work, not a version of this repository.

This is a conditional formalization, not an unconditional Lean proof of Beal's Conjecture. The compact referee map is:

B17_MazurPrimeCoverage.lean has no unrestricted fallback. The p = 13 branch is forced through Frey_13_exclusion_of_global_certificate; the other seven primes are separate fields. The constructor therefore proves that complete supplied evidence inhabits MazurRationalIsogenyClassification, but the repository does not manufacture the still-missing mathematical certificates.

The active conditional chain is:

primitive Beal data → canonical Frey model and exact rational E[2] → complete supplied B17 isogeny exclusions → externally supplied global conductor and 2-adic exponent-one certificate → Wiles form token and quotient plan → Tate-derived odd-prime exact divisibility → explicit EnrichedPlanSupplier carrying the Galois, Hecke, Eutheos, Taylor–Wiles, support, and transport data → level 2 → $S_2(\Gamma_0(2))=0$.

Closing the B17 case split does not close the independent Wiles, Tate, Ribet, patching, or conductor boundaries. The final B20 theorem still uses exactly the three named domain interfaces wiles_modularity, frey_conductor_data, and tate_step2_odd_prime_external, plus the explicit TwoAdicExponentOneSupplier and EnrichedPlanSupplier. Foundational audit entries such as propext, Classical.choice, and Quot.sound are Lean foundations, not additional mathematical theorems.

The v10.0.0 final assembly is lean/Beal/Final/ConditionalBealTheorem.lean. It composes the reproducible v9.2--v9.4 level-26 chain with the existing Frey, modularity, and Ribet level-lowering interfaces to prove the public Lean proposition BealConjecture.

This is a conditional theorem. Its mathematical boundaries are visible theorem arguments; the release does not claim that transcripts or JSON files construct the missing geometry.

The release chain, with its historical anchors, is:

• v9.2.0: rank 0, with the recorded Selmer conclusion {1} and the explicit second-descent soundness boundary;
• v9.3.0: merge b06254c, with dim J0(26) = 2 = 1 + 1 and the displayed isogeny;
• v9.4.0: documentation/verification commits e665e0d and e3e1e84, with M3 rank 2, the formal-immersion bridge, DOI badge, and CI run 33652336887; and
• v10.0.0: ConditionalBealTheorem, with five explicit premises and no new global axioms.

The five explicit premises are:

• J0DecompositionSoundness_26 J0_26;
• MwrankCertificateSoundness_26;
• FormalImmersionSoundness_26 J0_26 cotangent;
• FreyCurveExists; and
• LevelLowering_26.

The first three are the v9.2--v9.4 certificate-interpretation boundaries. FreyCurveExists reuses the existing level-26 Frey construction supplier; LevelLowering_26 carries the indexed modularity and Ribet level-lowering data. None is a global Lean axiom. Task #495 remains the named v10.0.1 hardening step for constructing the actual cotangent map behind M3.

The v9.4.0 release adds a reproducible finite formal-immersion witness at the prime 2. Its source triad is:

• sagemath/formal_immersion_26.log, the archived matrix transcript;
• sagemath/GENUINE_FORMAL_IMMERSION_26_CERT_v9.4.0.txt, the six-point evidence checklist; and
• lean/Beal/Mazur/Gates/FormalImmersion_26_Cert.lean, which houses the kernel-checked M3 data and rank certificates.

The witness checks that the displayed 2 ×ばつ 6 matrix has rank 2 over both Q and GF(2), matching the verified dimension dim J0(26) = 2.

Verification passed:

• formal-immersion JSON witness;
• existing J0(26) JSON witness;
• focused Lean build of Beal.Mazur.Gates.FormalImmersion_26;
• full main CI; and
• axiom audit with only the standard Lean foundations propext, Classical.choice, and Quot.sound.

The three soundness inputs remain explicit proposition-valued arguments: J0DecompositionSoundness_26, MwrankCertificateSoundness_26, and FormalImmersionSoundness_26. They are not global axioms. In particular, the finite matrix rank does not by itself construct the scheme-level cotangent map or prove Mazur's criterion; those geometric interpretations remain visible at the formal boundary.

The current conditional level-26 chain is:

v9.2.0 rank 0 (Selmer = {1}) → v9.3.0 dim J0(26) = 2 = 1 + 1 isogeny → v9.4.0 M3 rank 2 formal immersion → X0(26)(Q) finite.

The v8.9.0 release is the first honest executable Phase A computation for the level-26 route. Mazur/Gates/SecondDescent_Real_26.lean evaluates the complete finite grid of eight S-unit representatives against ten quartics at the two bad primes, 2 and 13: 8 ×ばつ 10 = 80 entries. The existing finite-field result tables are checked by decide, and all eighty entries pass both local checks.

This result is intentionally not presented as a singleton 2-Selmer calculation. The available evaluator is independent of the S-unit index, so the audit retains all eight representatives and proves that its candidate set is not a singleton. A future genuine singleton theorem must supply an S-unit-dependent covering map and a verified local obstruction table; filtering to the identity representative would be circular.

The release also moves the eight-element S-unit index and ten-row ledger into Descent_26_PhaseA.lean, preserves SecondDescentHypothesis_26 as an explicit conditional boundary, and adds ConditionalBealTheorem.lean as the aggregator for the conditional Phase B+C+D chain. The exact focused builds pass:

lake build Beal.Mazur.Gates.SecondDescent_Real_26
lake build Beal.ConditionalBealTheorem

The audited declarations contain no sorry, admit, sorryAx, or Lean.ofReduceBool; their remaining Lean foundation dependencies are propext, Classical.choice, and Quot.sound. This remains conditional mathematics, not an unconditional proof of Beal's Conjecture.

The v9.1.0 release adds the first real finite Phase C matrix evidence to the level-26 chain. The implementation was merged on main at 0a28f246359f1eea99a9183420a9fc395e03d019; the annotated release tag is created on the subsequent documentation-inclusive snapshot so that the tag, README, and source code describe the same release.

Mazur/Jacobian/FormalImmersion_26.lean now records:

• the explicit hyperelliptic model $y^2 = X0_26_f6(x)$, reusing the verified degree-six level-26 model;
• the finite differential basis $\omega_1 = dx/y$ and $\omega_2 = x,dx/y$;
• mod-3 reduction tokens for all four cusp labels, finite Abel–Jacobi replay rows, and two selected distinct cusp reductions;
• the differential evaluation table over $\mathbb Z/3\mathbb Z$; and
• the computed matrix $$ M_3 = \begin{pmatrix}1&10円&2\end{pmatrix}, \qquad \det(M_3)=2\ne 0\pmod 3. $$

The declaration formal_immersion_matrix_rank_two_decided proves the rank-two finite certificate by decide, without additional axioms. The combined certificate is exposed as FormalImmersionAt3_Real_26 and formal_immersion_real_evidence, and is exported through ConditionalBealTheorem.lean as phase_c_formal_immersion_evidence_real.

The focused release builds are:

lake build Beal.Mazur.Jacobian.FormalImmersion_26
lake build Beal.Mazur.Jacobian.J0_26_Decomp
lake build Beal.Mazur.Gates.SecondDescent_Real_26
lake build Beal.ConditionalBealTheorem
lake build Beal

This is real finite matrix evidence, not a claim that Lean has reconstructed the scheme-level Abel–Jacobi map, smooth reduction, or the geometric formal-immersion implication. Those interfaces remain explicit conditional data in FormalImmersionAt3_26. The genus-two Jacobian isogeny $J_0(26)\sim E_{26a1}\times E_{26b1}$, modularity, level-lowering, and the global four-cusp conclusion therefore remain conditional boundaries.

The v9.1.0 chain is:

real 80-check Phase A → real $S_2(26)$ q-expansion, Hecke, distinctness, and dimension-two Phase B evidence → real $M_3=\left(\begin{smallmatrix}1&10円&2\end{smallmatrix}\right)$ rank-two mod-3 Phase C certificate → conditional Phase D Beal endgame.

The Phase A and Phase B evidence remains as documented below. In particular, the 80-entry audit does not claim a singleton Selmer group, and the Phase B finite tables do not by themselves prove the missing Jacobian/isogeny theorem.

The v9.3.0 release replaces the external J0(26) decomposition trust boundary with a reproducible certificate. The committed Sage transcript and JSON witness record:

• $\dim J_0(26)=2=1+1$;
• the reported isogeny $J_0(26)\sim E_{26a1}\times E_{26b1}$;
• the exact models $(1,0,1,-5,-8)$ and $(1,-1,1,-3,3)$; and
• the explicit source-to-Lean factor correspondence, including its intentionally preserved factor-order difference.

The Lean bridge exposes J0DecompositionSoundness_26 and MwrankCertificateSoundness_26 as visible proposition-valued arguments, not global axioms. The finite model, dimension, transcript, and certificate integrity checks are kernel-checked; Sage decomposition semantics and mwrank/second-descent rank semantics remain explicit mathematical premises.

Verification for this release:

python3 scripts/j0_26_decomp_certificate.py --check
lake build Beal.Mazur.Gates.J0_26_Decomp
Main CI passed in 2m42s

The JSON witness check and focused Lean build passed, and no new axiom, sorry, or admit was introduced. This is reproducible evidence for the level-26 bridge, not an unconditional proof of Beal's Conjecture or of the still-missing global $X_0(26)(\mathbb Q)$ exhaustiveness argument.

The v9.2.0 evidence records the finite local checks and the explicit Selmer-to-rank boundary for the level-26 route. Its Sage-facing artifacts and Lean endpoint are documented in sagemath/README.md and SecondDescent_Singleton_26_Reproducible.

The rank-zero conclusion is retained as an explicit certificate premise in the v9.3.0 and v9.4.0 bridges; it is not silently promoted into an unconditional Selmer theorem.

The immutable release archive beal-conjecture-v8.9.0.tar.gz is 175,208 bytes with SHA-256 61362de15bd0c2ea9f64fea47ce91e958c9eafd2a6aa034147fda68099266672. The same bytes are attached to the GitHub release and Zenodo v8.9.0.

The v8.8.0 release added Mazur/Frey/LevelLowering_26.lean. Its proof-relevant interface carries a primitive Beal counterexample through explicit Frey-construction, modularity, level-lowering, and displayed-model boundaries to a noncuspidal rational point on the level-26 model. The existing Phase A–C rank-zero and four-cusp certificates then yield the conditional contradiction.

This is a conditional formalization milestone, not an unconditional proof of Beal's Conjecture. The new principal theorems compile with no sorryAx and use only Lean's standard {propext, Classical.choice, Quot.sound} foundations.

The v7.3.0 bridge makes the old/new boundary more explicit without pretending that fixed-point arithmetic has reconstructed modular-form geometry:

• The Eutheos certificate carries the denominator-cleared fixed-point inequality and list anchor as choice-free data.
• EutheosGeometryInterface supplies named old and new submodules, typed degeneracy maps αp and βp, their exact joint old-image representation, Hecke stability, genuine-form generation, V-membership, and coverage V = old + new.
• separation_kernel (j : EutheosJitter) is the lower-level old/new intersection statement old ∩ new = ⊥, explicitly parameterized by the jitter witness and its inequality.
• OldNewDecompHyp_from_Eutheos constructs the existing OldNewDecompHyp only at the theorem boundary from those fields. The arithmetic inequality alone is not claimed to prove the geometric kernel.
• B15 derives its old/new witness at the call site. In v8.0.0, the former hRank proposition is replaced by transparent TaylorWilesPatchingData.

The new supplier and B15 edge audit to [propext, Quot.sound], with no sorryAx, opaque declaration, or new mathematical axiom. The real-number bridge remains separately audited as [propext, Classical.choice, Quot.sound].

At the #161 baseline, the README described the formal boundary as one explicit modularity_hypothesis: a single typed proposition standing for the modularity portion of the argument. That was a useful first boundary, but it hid three mathematically different obligations behind one name.

The B14–B20 path now exposes the step boundary as data:

The distinction matters. tate_frey_multiplicative_derived is a theorem, not another axiom: it proves unit $c_4$ and bad discriminant for the canonical Frey model, then applies the narrow external Tate theorem through tate_step2_I_n_conductor_one. The model stores the Beal equation itself, so its discriminant is certified to equal the expression computed from the displayed Weierstrass coefficients. The canonical coefficient model is constructed without Classical.choice; global conductor data and the local Tate theorem are separate named boundaries. B15 proves that a certified plan repeatedly transports the token and that the terminal token contradicts $S_2(\Gamma_0(2)) = 0$.

This is deliberately a typed scaffold, not a claim that Wiles's theorem alone produces the certified descent plan or that this repository constructs modular forms. Those global ingredients remain inside the explicit wiles_modularity boundary until they are formalized.

• #161–#165: the Galois, Frey, Beal, and $S_2(Γ_0(2))$ scaffolding was consolidated into a typed Core/Wrapper architecture.
• #165: the Frey/Tate boundary introduced named Wiles, Tate, and Ribet interfaces instead of leaving modularity as one undifferentiated gap.
• #190–#198: the Frey curve's $c_4$, discriminant, conductor, local nonvanishing, and Tate-derived conductor witness were connected into a machine-checked per-prime statement.
• v7.0.0-genuine-provider: the opaque ribet_single_step : Prop was deleted from the active source path. B15 now consumes explicit RibetSingleStepProviders, and ribet_single_step_from_genuine audits to [propext, Quot.sound].
• v7.2.0: B15 consumes EnrichedPlanSupplier. Each indexed edge derives its q-expansion premise from normalized eigenline data and carries explicit old/new, rank, newform-support, and token-transport data. The former broad provider is no longer used by the active descent.
• v7.3.0 Task #440: B15 derives OldNewDecompHyp from the typed EutheosGeometryInterface at the use site. The supplier exposes the degeneracy maps, exact old-image representation, Hecke-stable old/new modules, genuine generation, V-coverage, and the jitter-indexed separation kernel. Only hRank remains as the explicit proposition-valued edge boundary; the geometric kernel remains an explicit supplier input.
• v8.0.0: B15 replaces that hRank proposition with TaylorWilesPatchingData. Finite generators and coordinates are coherent under transitions; the localized coordinate is derived through the level-zero projection.
• Patching specialization refinement: the level-zero reconstruction law is now proved from transition-coordinate compatibility, finite-level reconstruction, and projection extensionality rather than stored as a field. The patching data is indexed by the exact Frey residual representation and its exact Hecke attachment. It carries explicit deformation-to-localized-Hecke and finite-level diamond comparison data. The localized residual map extends the attachment evaluation, but no universal deformation property, Frobenius control, or R=T isomorphism is claimed.

The repository is organized into three layers, with a separate boundary for the named mathematical results used by the final theorem.

The Cores are the mathematical skeleton. Each states a real proposition in pure Lean, with zero imports from Mathlib and zero axioms beyond propext. They cannot hide sorry behind a library import. CI enforces this with #print axioms on every Core declaration.

Wrappers translate Core statements into the concrete Mathlib setting — real numbers, field norms, and GCDMonoid. They may carry foundational Lean dependencies such as propext, Classical.choice, and Quot.sound; they must never introduce sorryAx. The real-number BSD/Hasse boundary has its own audit.

The final B20 proof is conditional on three named mathematical boundaries and two explicit data suppliers:

1. Global conductor: frey_conductor_data supplies one conductor and its prime support for the canonical integral Frey model.
2. Wiles: wiles_modularity supplies a residual prime, a typed form token, and an odd-prime quotient plan for that fixed conductor.
3. Tate: tate_frey_multiplicative_at_model derives p ∣ N and p2 ∤ N at every intermediate edge from the fixed conductor, its prime support, and the narrow tate_step2_odd_prime_external boundary.
4. Genuine plan: EnrichedPlanSupplier enriches that Tate-certified plan. Each GaloisEdgeWitness carries the 07g–07k representation, Hecke, patching, support, and token-transport data; it no longer carries a duplicate exact-divisibility proof.

The current development implements the architecture below. Three named interfaces, an explicit 2-adic exponent-one supplier, and an explicit enriched-plan supplier mark the remaining inputs; the connective arithmetic around them is explicit Lean code.

For a primitive Beal triple $(A^x, B^y, C^z)$, the development forms the Frey curve $$ E : y^2 = x(x - A^x)(x + B^y). $$ B14_FreyTate.lean defines its $b_2$, $b_4$, $c_4$, discriminant, and model-indexed conductor data. It also proves that, under the stored Beal equation, the discriminant formula equals the expression computed directly from the Weierstrass coefficients. B14_TateC4Nonzero.lean proves the needed local $c_4$ nonvanishing, and B14_TateInImpliesOrd1.lean proves the unit-$c_4$ and bad-fiber certificates, then combines them with the narrowly named external Tate theorem to prove tate_frey_multiplicative_at_model; an existential compatibility wrapper is retained as tate_frey_multiplicative_derived.

This derived theorem is not another assumption: it is the explicit bridge from the generic odd-prime Tate statement to the actual Frey model and its fixed conductor data.

The prime 2 is accounted for separately and more cautiously. two_dvd_frey_discriminant and two_dvd_frey_c4 prove the elementary divisibility facts for the canonical equation, yielding the literal discEvenC4Even invariant witness—not a reduction-type classification. FreyTwoAdicLocalData separately requires an explicit diagonal-scaling minimality certificate. External 2-adic work must supply a visible FreyTwoAdicConductorCertificate containing a specific exponent e and the proof 2^e ∣ N ∧ 2^(e+1) ∤ N for the same model's supplied conductor N. No certificate is constructed here, and no full minimal-model theorem, Kodaira classification, or complete 2-adic Tate algorithm is claimed. The final level-2 route explicitly asks for the stronger FreyTwoAdicExponentOneCertificate; it does not attribute e = 1 to Wiles.

The global FreyConductorData contract itself supplies an opaque natural number and prime support. It is named "conductor" as the mathematical boundary, but this repository does not internally construct or identify it with Mathlib elliptic-curve conductor data.

Tate selects a fixed FreyCurveModel. Given an explicit exponent-one certificate for that model's 2-adic conductor, wiles_modularity supplies a typed form token plus an odd-prime quotient path from its conductor to level 2. B15_RibetIterate.lean derives every exact-divisibility edge from the model-indexed Tate theorem and then passes the certified plan to an explicit EnrichedPlanSupplier. Every corresponding edge carries typed Eutheos geometry, exact residual-representation and Hecke-attachment indices, Taylor–Wiles patching data, localized Hecke data, the 07j support bridge, and explicit token-transport data. ribet_single_step_from_genuine derives support and constructs the target token. B16_BealFinal.lean then eliminates the terminal token using the verified equality $S_2(\Gamma_0(2)) = 0$.

The space $S_2(Γ_0(2))$ has dimension zero: no cusp forms of weight 2 and level 2 exist. This fact is verified in the repository by rfl. Once the Ribet chain reaches level 2, the hypothetical primitive Beal triple yields a contradiction.

"Green" means the Lean source elaborates, its declared dependencies are visible, and the audit checks pass. It does not mean that Wiles's, Tate's, or Ribet's theorems have been reconstructed from first principles.

CI enforces the boundary on every push and pull request:

• Build all bricks — all B01–B21 Cores and Wrappers compile
• Clean locked rebuild — a separate job removes generated .lake state and rebuilds from the committed lean-toolchain, lakefile.lean, and exact revisions in lake-manifest.json; it fails if the clean build cannot reproduce the project or changes either committed Lake input
• Check proof placeholders — exactly two localized level-26 gate experiments retain one sorry wall each; every other Lean source is rejected if sorry, sorryAx, or admit appears
• Reject trivial Core stubs — no : Prop := True or equivalent placeholder
• Audit every Core declaration — Cores remain import-free and zero-axiom
• Audit the real-number transport boundary — it may use Lean foundations but may not use sorryAx
• Audit final B20 declarations — #print axioms must expose wiles_modularity, frey_conductor_data, and tate_step2_odd_prime_external; a separate elaboration check requires the explicit TwoAdicExponentOneSupplier and EnrichedPlanSupplier parameters

The project uses Lean 4.12.0 and Mathlib:

export PATH="$HOME/.elan/bin:$PATH"
lake exe cache get
lake build Beal

Keep the same checkout and its ignored .lake/ directory between commits. Lake then rebuilds only changed project modules. CI caches the Lean toolchain and .lake/packages under a key derived from lean-toolchain and lake-manifest.json; lake exe cache get runs only when that exact dependency cache is missing. Avoid lake clean unless a genuinely clean rebuild is required.

CI also runs a release-level clean-checkout validation. That job deliberately removes generated .lake/ state and does not restore the incremental dependency cache before running lake build +Beal.B00_OperaNumerorum Beal. It must fetch and build using only the Lean version in lean-toolchain, the package declaration in lakefile.lean, and the exact transitive revisions recorded in lake-manifest.json. It then verifies that Lake did not rewrite either committed input. A failure means that the release snapshot is not reproducible from its committed dependency lock.

GitHub Actions repeats this clean-checkout release-snapshot check every Monday at 06:17 UTC, even when no code has changed. The scheduled run installs the exact toolchain named by lean-toolchain before Lake fetches the exact transitive revisions in lake-manifest.json. If either the toolchain or a locked Git revision disappears, the job reports that locked input explicitly instead of silently updating the snapshot.

Concept DOI (all versions): DOI

Zenodo version records preserve release snapshots; the Lean axiom report records what the snapshot still assumes. Green CI means that these dependencies are visible and checked. It does not mean that Lean has reconstructed Wiles, Tate, or Ribet from first principles.

The v4.0.0 Zenodo landing page retains an older v0.4-style display title; its release tag, archive, and audited boundary are the v4.0.0 row above. The

v4.1.0 — Tate truthful. The Tate boundary now carries one fixed FreyCurveModel for each fixed solution tuple (A, B, C, x, y, z). That model ties the Frey coefficients, c4, discriminant, prime support, and conductor together, so the conductor N is no longer an unrelated or freely chosen natural number. The closed B14 arithmetic establishes the Frey c4 identity and the divisibility facts used by the local argument without introducing another named mathematical axiom. For an odd prime p at which the model is minimal and the discriminant is bad, p ∤ c4, p ∣ Δ, the model contract yields p ∣ N, p2 ∤ N. Thus the derived conductor statement reuses the conductor of that same fixed Frey model and records local conductor exponent one at p.

This historical release formalized a typed odd-prime local Tate boundary. It did not claim a complete Tate algorithm, a formal 2-adic conductor calculation, or a global computation of N from the Frey coefficients. Its version DOI is 10.5281/zenodo.22091549.

When discussing an audit, cite the specific version DOI rather than the concept DOI. For example:

Fox, David. (2026). Beal Conjecture — Typed Preserved-Form Ribet Descent — v5.0.0. Zenodo. https://doi.org/10.5281/zenodo.22090900

• Andrew Beal (1997) — the Beal Conjecture.
• Gerhard Frey (1986) — the Frey curve and the bridge from Diophantine equations to elliptic curves.
• John Tate (1975) — algorithm for computing the conductor of an elliptic curve at a bad prime.
• Kenneth Ribet (1990) — level-lowering and the $\varepsilon$-conjecture.
• Barry Mazur (1978) — irreducibility phenomena for Galois representations.
• Andrew Wiles (1995) — modularity of semistable elliptic curves and Fermat's Last Theorem.
• Richard Taylor and Andrew Wiles (1995) — the ring-theoretic criterion completing modularity.

The audit ladder is intentionally scoped and chronological:

• v7.2.0-1419-infra (a457c8b558, CI 33139424482) closed the fixed-point jitter infrastructure and left two proposition-valued edges.
• v7.3.0 Task #440 (PR #3, CI #239, 53m23s, SUCCESS) closed the typed Eutheos old/new edge and left one explicit hRank proposition at that historical snapshot; v8.0.0 replaces it with patching data.
• v7.3.1 is this documentation patch: it archives the distinction between the clean focused bridge and the two intentional repository-wide boundaries.

The phrase "0 axiom" is scoped to the typed Eutheos focused boundary. It means that Beal.Galois.OldNewDecompHyp_from_Eutheos, EutheosJitter, and the jitter-indexed separation kernel introduce no proposition-valued axiom or opaque declaration of their own and audit to [propext, Quot.sound]. The V-specific edge is clean. The Task #440 build covered 2,424 targets and CI #239 completed successfully in 53m23s.

This is not a repository-wide claim. A literal grep -r '^axiom' over the repository returns exactly two intentional mathematical boundaries: mazur_irreducibility_axiom and wiles_lifting_axiom. They belong to the existing Mazur/Wiles B20 modularity tower and are not part of the typed Eutheos bridge. The repository-wide focused audit therefore distinguishes these named boundaries from the clean Eutheos supplier rather than hiding them.

The real-number interpretation of the fixed-point inequality ‖p·α0‖ < 1/p remains isolated in the desert-brothers module and audits to [propext, Classical.choice, Quot.sound]. That choice-bearing real bridge does not leak into the typed Eutheos/Beal path.

B15 no longer stores hRank : LocalizedRankOne .... Each edge instead carries finite prime levels Qn, finite patched modules, coherent transition maps, diamond operators, compatible inverse-limit projections, depth equalities, and a separately named specialization boundary. The localized coordinate is derived through the level-zero projection; it is not stored as a field. LocalizedRankOne_from_Patching then constructs the linear equivalence and audits to [propext, Quot.sound].

This is a refinement of the boundary, not a completed Taylor–Wiles proof. The Eutheos jitter inequality does not construct Taylor–Wiles primes, and the historical B05 Mazur/Wiles axioms do not state R=T, freeness, localization, or multiplicity one. The reconstruction law in PatchingSpecializationData names that missing R=T/localization step explicitly. Constructing the complete TaylorWilesPatchingData remains a Type-valued obligation of the enriched descent-plan supplier.

This repository is one chamber in Opera Numerorum, a connected Lean development whose rooms share a method rather than a claim of finished mathematics: state the arithmetic explicitly, make the connective steps machine-checkable, and leave every imported theorem or missing construction visible at the boundary.

The Beal route begins with primitive divisibility, builds the Frey curve, and then passes through conductor data, Galois representations, modular forms, level lowering, and the terminal fact that S2(Γ0(2)) = 0. The current level-26 work adds finite, reproducible evidence around the residual prime 13 while keeping the global Jacobian, Selmer, Tate, Wiles, and Ribet steps conditional where they are not yet constructed.

The v9.2.0 documentation and source are organized as a small audit trail:

• lean/ contains the numbered formal development;
• lean/Beal/Mazur/ contains the level-26 route;
• certificates/ contains committed certificate data;
• sagemath/ contains reproducibility transcripts and replay programs; and
• docs/ contains model, rational-point, and release documentation.

"Green" means that the stated Lean programs and audits elaborate. It does not mean that Beal's Conjecture has been proved unconditionally. The project records that distinction as part of the mathematics.

The Beal development is not itself a proof of the Riemann Hypothesis. It is one chamber of Opera Numerorum, a wider program in which different mathematical languages approach the same landscape: Arakelov geometry, automorphic forms, spectral gaps, arithmetic dynamics, and the analytic behavior of zeta functions. These related repositories are kept at the bottom so the formal Beal account above remains self-contained.

riemann-arakelov-positivity turns positivity on the modular curve $X_0(143)$ into an arithmetic inequality. Its architecture centers on $g(X_0(143)) = 13$, $\omega^2 = 48/13$, and the finite set $S_4 = {2, 3, 19, 191}$.

arakelov-rh-descent approaches the same territory through a spectral gap on $X_0(143)$, with $\lambda_1 \geq 975/4096$ as the Kim–Sarnak input.

rh-growth-contradiction takes a contradiction route, comparing growth permitted by a proposed zeta bound with Littlewood's $\Omega$-phenomenon.

brothers-desert-proof is a synthetic route studying zero repulsion directly, through the $p = 5$ bridge and the desert property of exceptional primes.

ImperialCollegeLondon/FLT is an inspiration for formalization, not a dependency of this repository.

The active companion to the conditionally complete v11.0.0 assembly is beal-level-26-foundations v1.2.1. The parent pins its immutable commit and imports the ledger-derived identity P * C3 = M3 = [[1,1],[0,2]] and finite formal-coordinate calculation rather than duplicating them. The actual Picard/Abel--Jacobi compatibility, genuine Selmer comparison, Tate/Kodaira conductor classification, and geometric Riemann--Hurwitz bridge remain explicit boundaries.

Maintained by DavidFox998 as part of Opera Numerorum: mathematics made auditable, with the beauty left visible.