# Governing Twin audit: Phase 2

Audit date: 2026-09-08. Scope: all 40 Markdown files in `bsd_source/phase2/files`, all 40 accompanying article HTML pages, and `phase2/index.html`. Inputs were read without modification. All article descriptions, claim boxes, chronological tags, and index cards were inspected; they are additional editorial assertions, not duplicate proof text. HTML bodies load the corresponding Markdown dynamically and are not self-contained theorem manuscripts.

## Verdict and scope

**CONCUR, with dependencies retained:** the final 696.e1 prime-family router is a plausible derived consequence of the cited primary theorems; the main theorem hypotheses checked below match its claimed reduction types. This audit does **not** independently certify its all-prime residual-image metadata, modular-symbol L-value, Manin constant, or reported 10^7-prime sweep. The final explicit theorem remains **UNVERIFIED HERE AS A COMPLETE ARITHMETIC INSTANTIATION**, rather than disproved.

**CHALLENGE:** the universal H3 claim in article HTML 09 and the Phase 2 index is false as written. With $W_-(E)=\{29\}$ and valuation $1$, the gcd is $1$ but $p=29$ has no distinct nonsplit witness. The Markdown qualifies its displayed universal claim by restricting to good supersingular/fixed additive branches; the HTML drops this essential qualification. The later leave-one-out network formulation repairs the issue. The final supersingular branch for 696.e1 is unaffected because its primes are good and hence distinct from 29.

This entire phase concerns analytic-rank-zero quadratic twists. It does not establish rank-two leading terms or close the general BSD conjecture. Prime density $1/24$ refers to the set of supporting primes, not density among all squarefree twisting integers.

## Dependency chronology

1. Main files 00–07 (HTML dated 2026-08-12) select a non-semistable extension of Banwait–Huang, initially seeking all odd primes through Fouquet–Wan. H1/H2/H3 are proposed compiler interfaces; these documents are plans, not closed theorem applications.
2. Main 08–13 (2026-08-13) derive H2 and H3 and discover that a universal FW-only ordinary route is unsuitable: ordinary primes with $a_p^2\equiv1\pmod p$ fail H2. The live architecture switches to a hybrid: existing ordinary/multiplicative/additive-twist theorems plus FW at supersingular/fixed additive primes.
3. Main 14–19 specialize to 696.e1, define a prime set of density $1/24$, and assert a provisional exhaustive router. The base BSD anchor is initially attributed too broadly to the Miller historical summary.
4. Main 20–28 replace that attribution by Creutz–Miller Theorem 1.1, claim successful referee audits, and upgrade to a derived theorem candidate. The submission gate nevertheless explicitly requires independent arithmetic/source reproduction. The archive has no actual independent referee transcripts or runnable 696.e1 arithmetic checker.
5. Main 29 is the formal 696.e1 theorem note; 30 abstracts it to a two-witness sufficient criterion; 31 permits arbitrary gcds, finite exceptional-prime routes, squarefree products, and fixed odd additive primes conditional on local and period certificates.
6. Side files 00–07 (HTML numbers 32–39, also dated 2026-08-13) extract gcd and period facts, then reduce H2 for a reducible local representation to a local isogeny/dual-kernel test. Their last item is an implementation specification. No actual implementation, local factorization certificate, or regression execution is included.

The website's `40/40 COMPLETE` counts documentation coverage. It is not mathematical closure. Its editorial dating also refers to posting work on 2026-08-18; that is distinct from manuscript dates.

## Primary-source checks

Primary PDFs and extracted text are saved under `work/twin_sources/`. URLs below identify exact originals; these are read-only downloads, not reproduced mathematics.

| Source | Version/date and pointer | Exact application checked | Audit conclusion |
|---|---|---|---|
| [Fouquet–Wan](https://arxiv.org/pdf/2107.13726v3) | v3, 2022-04-08; Theorem 1.1 p.3; Theorem 1.7 p.7; Corollary 1.10 pp.8–9 | Odd $p$; globally absolutely irreducible odd residual representation; forbidden local character ratio; ramified nonsplit Steinberg auxiliary prime distinct from $p$, exact conductor exponent one; nonzero central value for BSD corollary | Correct source for the supersingular route; a preprint, not a universal arbitrary-reduction statement without hypotheses. Period comparison remains additional. See normalization note below. |
| [Skinner](https://arxiv.org/pdf/1407.1093v1) | v1, 2014-07-04; Theorem C p.3 (published Pacific J. Math. 283 (2016), 171–200) | Good ordinary or multiplicative at $p\ge3$; irreducible $E[p]$; distinct multiplicative prime with ramified residual representation; nonzero central value | The source directly supports the claimed good-ordinary and multiplicative 3/29 routes, with Néron period. No global semistability hypothesis. The extra L-invariant clause in Theorem B is discharged for elliptic curves in Theorem C. |
| [BSTW](https://arxiv.org/pdf/2409.01350v2) | v2, 2024-09-11; Theorem 9.21(c) p.84; proof pp.84–85 | $p\nmid6N$, good ordinary, residual irreducibility; integral equality needs a residual ramification witness. The ramified quadratic-twist clause requires a witness away from the twisting discriminant | Supports the $p\mid d$ route with witness 29 (or a valid network witness). A subsequent rank-zero descent is needed; it is explicitly supplied by Banwait–Huang's proof. Integral equality must not be confused with equality only after inverting $p$. |
| [Banwait–Huang](https://arxiv.org/html/2601.16044v3) | v3, 2026-06-04; Proposition 2.9(1), Remark 2.10, Theorem 2.14(1) | The proposition's semistability is used to generate the witness in the additive-twist branch; the remark permits imposing it directly. The 2-primary theorem needs an optimal rank-zero base, odd Manin constant, base BSD at 2, correct L-value valuation, twist inertness and bad-prime splitting | The archive's final use of the negative-discriminant/no-2-torsion branch is correct conditional on the base arithmetic. Source v3 explicitly gives nonvanishing as well as the 2-part. |
| [Zhai](https://arxiv.org/pdf/1409.0231v2) | v2, 2017-12-03; Theorems 1.1–1.2 p.2; standing assumptions pp.1–2 | Original negative-discriminant/no-2-torsion twist nonvanishing and 2-part result; odd Manin, optimality, inert support primes, base 2-BSD and splitting at all bad primes | Independently confirms the original source behind BH 2.14(1). Zhai uses the least positive real Néron period; for negative discriminant the real-component factor is one, so no hidden factor-two adjustment for 696.e1. |
| [Creutz–Miller](https://arxiv.org/pdf/1105.4018v2) | v2, 2012-09-17; Theorem 1.1 p.2 | Every $E/\mathbb Q$ of conductor $N<5000$ and analytic rank at most one satisfies full BSD | The archive's repair is correct, conditional on 696.e1's conductor and analytic rank. It does not follow from numerical analytic Sha alone. |
| [Česnavičius–Neururer–Saha](https://arxiv.org/pdf/1911.09446v3) | v3, 2022-11-02; introduction p.2; published JEMS 26 (2024), 573–637 | Reviews $p\nmid c_\phi$ at semistable primes and Edixhoven's $p\ge11$ exception for additive potentially ordinary II/III/IV | Archive's good-supersingular period argument and quoted large-additive sufficient subclass agree with the source. For the latter, the exact original input is Edixhoven 1991 Theorem 3, reviewed here, not a new theorem of CNS. Original Edixhoven proof was not separately read. |

Latest arXiv abstract records were checked for FW (v3) and BSTW (v2); the versions used by the archive remain the current arXiv versions returned on the audit date. BCS's optional ordinary-exception route is not instantiated on any exceptional prime in this archive; its exact `(im)` condition remains a source/compiler obligation rather than an inherited certificate.

### FW normalization detail

The actual weight-two source extension in Theorem 1.7 is

$$
0\longrightarrow\mu\longrightarrow\bar\rho\longrightarrow\mu\omega^{-1}\longrightarrow0,
$$

with nontrivial unramified quadratic $\mu$. The Tate-module convention for elliptic torsion at nonsplit multiplicative reduction is

$$
0\longrightarrow\mu\omega\longrightarrow E[p]\longrightarrow\mu\longrightarrow0.
$$

Thus one must explicitly use $\bar\rho=E[p](-1)$ when matching these formulas. H1 and the forbidden-ratio property H2 are preserved under a scalar character twist. This is a repairable convention omission in phrases such as “$E[p]$ satisfies Theorem 1.7,” not a counterexample to the nonsplit-witness rule. The source itself identifies its weight-two automorphic hypothesis with nonsplit multiplicative reduction.

### Local H2 lemmas

For $V=E[p]$ reducible over $\mathbb F_p$, write $V^{ss}=\lambda\oplus\mu$ with $\lambda\mu=\omega$. The archive's equivalence between the forbidden ratio and $\lambda^2=1$ or $\mu^2=1$ is valid. Its isogeny/dual test is also valid: the character on the dual kernel is the other constituent, and $x(P)$ rational is equivalent to all conjugates being $\pm P$. A characteristic-zero cyclic prime-order kernel polynomial has a linear factor exactly in that situation. The irreducible-over-$\mathbb F_p$ shortcut is sufficient here because every forbidden constituent forced by determinant comparison has values in $\{\pm1\}\subset\mathbb F_p$.

This proves an exact criterion, not the existence of a supplied backend that constructs the isogeny over $\mathbb Q_p$ and certifies its factorization. The final implementation remains missing. Potentially multiplicative local curves fail H2 as stated; absence of rational local $p$-torsion alone does not imply H2.

## Per-file coverage and live status

All entries below include their matching HTML article. `main` means the original 00–31 file numbering; `side` means the duplicate 00–07 side filenames displayed as HTML 32–39.

| Markdown file | HTML number | Coverage/status |
|---|---:|---|
| 00_Phase2_Global_Enclosure_Consensus.md | 00 | Read; route-selection history, superseded FW-only ambition. |
| 01_Phase2_Route_Matrix.md | 01 | Read; scope explicitly restricted to extensibility of rank-zero family theorem. |
| 02_Fouquet_Wan_Hypothesis_Compiler.md | 02 | Read; initial incomplete H1/H2/H3 interface, replaced by exact local notes. |
| 03_Quadratic_Twist_Invariance_Bridge.md | 03 | Read; character tensor/local splitting lemmas valid; no all-prime closure by themselves. |
| 04_Finite_Exceptional_Prime_Problem.md | 04 | Read; historical heuristic; absolute irreducibility vs rational isogeny wording requires odd-global-representation argument. |
| 05_NonSemistable_Family_Theorem_Schema.md | 05 | Read; explicitly conditional schema. |
| 06_Phase2_Agent_Experiment.md | 06 | Read; experiment plan only; no 60-curve execution assets present. |
| 07_Stop_Rules_and_Claim_Ladder.md | 07 | Read; methodological ladder, no theorem evidence. |
| 08_FW_Weight2_Exact_Translation.md | 08 | Read; H2 valid; H3 needs convention mapping above. HTML claim box also reverses PASS/FAIL wording, inconsistent with its own prose. |
| 09_FW_H3_Exact_Compiler.md | 09 | Read; valid only with the branch qualification excluding witness primes. HTML/index universal claim invalid. |
| 10_FW_H2_and_Ordinary_Obstruction.md | 10 | Read; ordinary congruence correct; key hybrid-route transition. |
| 11_Derived_Supersingular_FW_Bridge.md | 11 | Read; correct conditional derived bridge, period gate explicit. |
| 12_Hybrid_Odd_Prime_Router.md | 12 | Read; complete branch plan conditional on stated hypotheses. |
| 13_Candidate_NonSemistable_Strong_BSD_Family.md | 13 | Read; candidate with six open obligations. |
| 14_Candidate_Sieve.md | 14 | Read; 696.e1/control-116.b1 project arithmetic, not reproduced. |
| 15_696e1_Base_Certificate.md | 15 | Read; initial Miller attribution superseded by 21; analytic-Sha-to-exact-Lalg step unverified arithmetic. |
| 16_696e1_Chebotarev_Support.md | 16 | Read; field-theoretic density derivation valid conditional on cubic invariants; finite prime computations not replayed. |
| 17_696e1_All_Prime_Router.md | 17 | Read; correct exhaustive reduction partition; dependencies retained. |
| 18_Provisional_Derived_Theorem.md | 18 | Read; project CLOSED labels superseded by later audit status, not accepted as independent evidence. |
| 19_Independent_Referee_Handoff.md | 19 | Read; requested referee assignments do not establish their execution. |
| 20_Adversarial_Referee_Verdict.md | 20 | Read; claims prior referee success, actual transcript absent. |
| 21_Base_BSD_Anchor_Repair.md | 21 | Read; source repair independently verified in CM Theorem 1.1. |
| 22_Odd_Prime_Source_Audit.md | 22 | Read; theorem statements checked independently above. |
| 23_FW_Supersingular_Source_Audit.md | 23 | Read; primary FW checked; explicit Tate twist clarifies omitted normalization. |
| 24_Manin_Period_Audit.md | 24 | Read; good-prime support theorem correct; singleton isogeny-class input unverified here. |
| 25_Chebotarev_Referee_Audit.md | 25 | Read; same valid field argument, reported independent sweep absent. |
| 26_Novelty_Search_Log.md | 26 | Read; no-hit does not establish novelty; its search not independently reproduced. |
| 27_Revised_Derived_Theorem_Candidate.md | 27 | Read; final provisional router, arithmetic instantiation remains unverified here. |
| 28_Submission_Gate.md | 28 | Read; explicitly asks for missing arithmetic/source reproduction. |
| 29_Theorem_Note_v1.0.md | 29 | Read all 817 lines including references/Appendix A; claims formal result but appendix is a checklist, not executable certificate. |
| 30_Two_Witness_Criterion_v0.1.md | 30 | Read all 550 lines; sufficient criterion valid subject to source/normalization dependencies; T6 is universally quantified, not itself an executable finite check. |
| 31_Witness_Network_Criterion_v0.2.md | 31 | Read all 788 lines; gcd/LOO reduction valid, exceptional-prime and period routes explicitly conditional; finite prime set alone does not make twist-uniform certificates supplied. |
| 00_GCD_Witness_Lemmas.md | 32 | Read; elementary gcd/LOO facts valid. |
| 01_Odd_Additive_Period_Barrier.md | 33 | Read; correct period obstruction and reviewed sufficient subclass. |
| 02_Next_Compiler_Targets.md | 34 | Read; pending implementation/period/BCS tasks. |
| 03_FW_H2_Jordan_Holder_Lemma.md | 35 | Read; valid determinant/character lemma. |
| 04_Local_p_Isogeny_Kernel_Criterion.md | 36 | Read; valid local isogeny/dual test as abstract criterion. |
| 05_Kodaira_Prefilters_and_NoGo.md | 37 | Read; valid limited no-go shortcuts; no complete Kodaira-only classifier. |
| 06_Witness_Network_v03_Integration.md | 38 | Read; pseudocode integrates valid tests but no executable artifact. |
| 07_Local_Agent_Implementation_Spec.md | 39 | Read; specification and desired fixtures only, not executed tests. |

## Missing assets and unclosed claims

- No accompanying executable 696.e1 checker, exact modular-symbol result, all-prime residual-image certificate, or machine-readable base certificate is present in Phase 2.
- The reported $27,667$ primes below $10^7$ and $a_{241}=-7$ are project assertions. No sweep code/output is present. They are unnecessary for the Chebotarev proof and were not recomputed here.
- The local H2 backend and fixtures A–F are specifications. No Sage/Magma session, local-isogeny coefficients, p-adic factorization or Hensel certificate exists in the supplied phase.
- The finite-exception theorem schema requires exception certificates for every twist in its family. Merely checking a fixed set of primes on the base does not automatically prove every proposed alternative theorem's image/period assumptions on all twists. For the core FW local conditions, the specified splitting makes them genuinely invariant.
- The base-specific strong BSD claim and reusable low-rank family schemas do not connect to the rank-two complex leading-term comparison by any proved edge in Phase 2.

## Targeted action for lead

Keep the correct 696.e1 conditional router as the strongest low-rank branch, flag the HTML H3 overquantification and missing reproducibility assets, and do not spend the main attack budget rerunning an irrelevant density sweep. The independent p5 normalization issue is a stronger candidate for a substantive correction. The local Bockstein construction discussed with the lead is mathematically sound but needs to be reported as a local cohomological realization; it does not by itself identify a global BKS/Nekovář Selmer regulator.