CHALLENGE (circularity): positivity_geometry

Hunt #150, mechanism 4. Circularity challenger, 2026-10-26, branch hunt/rh-moonshot-challenge-circularity. This session did not build the construction and edited none of its files. It read NOTES.md and frobenius_span.py as committed at 6c3b90e4b and scored them against the rubric pre-registered in MISSION.md (section "Challenge-phase scoring rubric", 2026-10-26).

Files here: circ_positivity.py (independent recomputation, shares no code with frobenius_span.py), test_positivity_geometry_challenge_circularity.py (8 tests, every number below), this page. The builder's own 58 tests pass in this checkout.

Grades as in MISSION.md. Nothing written here is kernel-checked.

Verdict

Unresolved, as the builder says; the restricted-class obstruction is downgraded to a scoped refutation of a loosely defined class. No hidden use of RH, GRH or the zeros was found in any definition or computation. The target H_Q is an RH-equivalent restatement, which the builder states openly and which is already in print (Connes 1999; Connes, Consani and Marcolli 2008), so it is recorded as a reformulation. Three defects, none fatal to the disposition:

• D1 (GRH in the scope boundary). NOTES section 5.4, lines 389-391, and section 9, lines 574-576, say a degree-1 Q-analogue of the golden fake "cannot exist", so the obstruction does not reach Q. Unconditionally, Kaczorowski-Perelli says only that a degree-1 element of the Selberg class is zeta or a shifted Dirichlet L-function; such an element violating RH exists if and only if GRH fails for some Dirichlet L-function. The scope boundary is therefore conditional on GRH and should be stated as such. The obstruction itself does not use it.
• D2 (the class is not explicitly defined). Section 9 calls the obstruction a restricted-class obstruction for arguments that "transfer verbatim to global function fields". ALIGNMENT.md section 5, item 2, asks for an explicitly defined class and a theorem whose hypotheses are shown to apply. "Transfers verbatim" is not a definition; what the golden fake actually establishes is a counterexample to one implication in the function-field world: {rationality, curve functional equation, curve pole structure, nonnegative integer closed-point counts in every degree, a ±1 local factor per place of F_5(t) with conductor 4} does not imply RH. That is disposition 1 (refuted) for that implication, which is sharper and fully supported; "restricted class obstructed" for number-field arguments overreaches by the vagueness of the transfer clause.
• D3 (prior art on the phenomenon). Ihara zeta functions of (q+1)-regular graphs carry an Euler product over primitive closed paths with nonnegative counts, a functional equation and rationality, and satisfy their RH analogue exactly when the graph is Ramanujan (Ihara 1966; Sunada; Hashimoto; Stark-Terras). For a non-Ramanujan graph the factor 1 - lambda u + q u^2 with |lambda| > 2 sqrt q has precisely the golden fake's shape (q = 5, lambda = 5 gives 1 - 5u + 5u^2). The Ihara zeta carries extra (1 - u^2)^(r-1) factors, so it does not have the curve pole structure, and its Euler product is over graph primes, not places of F_q(t). The exact form is therefore not covered, but "Euler product with nonnegative prime counts plus functional equation does not give RH" is published, and the NOTES' novelty search (section 12) missed it.

Scorecard

1. Disposition. Unresolved (agreeing with NOTES section 9), with the golden fake recorded as a refutation of the function-field implication listed under D2, and the "restricted class obstructed" label withdrawn to that scope (D2). Evidence: circ_positivity.quadratic_split, closed_points; NOTES lines 562-576.
2. Grade per step.
3. Shadow H_F on 8 curves, exact inertia: hardened (exact integer arithmetic, two inertia routes; builder's test_positivity_geometry.py, 58 passed here). Agreed.
4. H_Q equivalent to RH (NOTES 1.2): derived, self-reviewed; I rechecked the projection (P(gamma_f) orthogonal to F1 and F2, block diagonal) and the reduction to real f (the form W(f * g*) is symmetric because the zero set is closed under rho -> 1 - rho, so complex f splits into its real and imaginary parts). Agreed, and published: the grade of the reformulation is "cited" (Connes 1999; Connes-Consani-Marcolli 2008, arXiv math/0703392, section 6.2 "Weil positivity and the Riemann hypothesis").
5. Window-2g lemma (2.4): derived, self-reviewed. I rechecked the proof (t_k symmetric in k because the multiset is inversion-closed; the interpolation uses at most 2g nodes). Independent measured check on 120 random conjugation- and inversion-closed multisets of genus 2, 3, 4, on and off the circle, not Weil-type polynomials: 0 disagreements (test_window_2g_lemma_on_arbitrary_closed_multisets). Sharpness at genus 2 rebuilt by float eigenvalues: min eigenvalue >= 0 at window 3 and < -1e-3 at window 4. Grade stays derived, now with a second-session measured check.
6. Sharpness in random families (2.4): the on/off-circle classification is by mpmath roots at 40 digits, so the "on-circle cases positive on every window" half is measured, not hardened; the refusals are exact. Minor overgrade.
7. Golden fake in every degree (4.2, 5.3): derived for all degrees, self-reviewed, exact below thresholds. Recomputed to d = 300 here by separate code: b_d nonnegative integers, min b_d = 1; split and inert counts nonnegative integers to d = 300, first eight values equal to NOTES line 338-339. Ramification configurations {1: 4}, {1: 2, 2: 1}, {2: 2}, {4: 1} fail by d = 10 (planted control). Agreed.
8. Number-field G1/G2 table (6.2): measured. Lambda_DH(6) = (1 + kappa^2) log 6 recomputed by a separate log-derivative recursion to 25 digits. Agreed.
9. Composite: the obstruction is derived, self-reviewed, as NOTES says; the mechanism's RH step (G3, an ambient Hodge index) does not exist, so no grade attaches to it.
10. The Euler-product step. Named: G1, the Lefschetz identification of off-diagonal intersection numbers as point counts (NOTES 6.1). Tested on rivals that lack it: DH fails G1 (Lambda(3) < 0, Lambda(6) != 0), Epstein (1,1,6) fails at n = 48 and n = 6. The builder correctly says G1 is a precondition and not the engine; G3 uses no Euler product and is missing for zeta.
11. Finite stand-ins. None found standing in for a uniform step in a claim. The all-degree golden-fake statements carry derived tail thresholds (D0 = 2, D1 = 4) with exact checks below them; the 8-curve shadow is presented as a finite exact check of a theorem (Weil), not as evidence for zeta. The window-2g statement is derived, not scanned. The G1 table runs to n = 60 and is labelled measured, and its role is only to exhibit failures (a finite range suffices for a failure).
12. Novelty search. Recorded by the builder (two web searches, NOTES 12). This session's queries, 2026-10-26, web search: (a) "Ihara zeta function regular graph Riemann hypothesis Ramanujan non-Ramanujan functional equation Euler product counterexample": found Sunada's equivalence (RH for Ihara zeta iff Ramanujan), see D3; (b) "Connes Consani Marcolli Weil proof adeles class space Hodge index ... equivalent Riemann hypothesis": found arXiv math/0703392, where RH is equivalent to positivity of a trace pairing, which is H_Q's content; (c) "function field analogue Selberg class zeta functions ... not satisfying Riemann hypothesis F_q(t)": nothing on golden-type fakes over F_q(t); (d) "Weil explicit formula function field positivity finitely many test functions ... Toeplitz 2g": nothing explicit on the window-2g statement (Carathéodory- Toeplitz-Pisarenko is the likely home, not confirmed). arXiv, zbMATH and MathSciNet full text, and Lean Zulip, were not searched by this session.
13. Rivals on complete hypotheses. The mechanism's complete hypotheses for zeta do not exist (G3 is missing), so no rival can refute an RH implication here; the entries concern the stated preconditions and the restricted-class claim.
14. DH: lacks G1 and G2 (no pole; signed, off-support Lambda). Not that counterexample. Measured, n <= 6 suffices.
15. Epstein (1,1,6): lacks G1 (first negative Lambda at 48, off-support at 6; builder's measurement, n <= 60). Not that counterexample.
16. W_{1/4}: lacks G2 (two poles, signature (2, 2)). Not that counterexample.
17. zeta(s - delta), delta = 1/10 and 1/4 (computed here, not in the NOTES): passes G1 and G2 (Lambda(n) n^delta nonnegative and on prime powers for n <= 60; one pole at 1 + delta, residue 1) and violates RH (zeros at 1/2 + delta + i gamma). It lacks the functional equation about Re s = 1/2, which the restricted class includes, so it is not a counterexample to the golden-fake class. It does show over Q, without the function-field detour, that G1 and G2 alone separate nothing, which strengthens the builder's "preconditions, not engine". Measured, n <= 60.
18. Golden fake (function field): meets every listed function-field hypothesis and violates RH: refutes that implication (D2). Derived, self-reviewed, with this session's exact recomputation to d = 300.
19. Circularity. No definition reads the zeros. Requirement A provenance, by hand: frobenius_span.py imports numpy, mpmath, flint, sympy and, inside the two rival routines nf_log_derivative_profile and nf_pole_residues only, zeta.epstein's coefficient interfaces and function values near s = 1; nothing imports zetazero, siegelz, a zero table or a xi-phase. Point counts are literal enumeration (count_points), and the Weil-polynomial roots appear only as a cross-check (eigenvalue_moduli_ratio). W is defined in NOTES 1.2 as a sum over zeros, but the pairing is evaluated through the explicit formula (the pole terms cancel exactly and leave the prime and archimedean sides), an unconditional identity, so that is not zero-importing. The only conditional input is D1: the claim that the obstruction does not reach Q rests on GRH for Dirichlet L-functions.

What this challenge does not settle

The rival-gate challenger's computation is separate. Whether some square of Spec Z carries a pairing that passes G1 and G2 (NOTES section 11, step 1) is untouched here.

Run manifest

id: rh_moonshot-positivity_geometry-2026-10-26-challenge-circularity
hunt: rh_moonshot
started: 2026-10-26T12:20+00:00
finished: 2026-10-26T13:20+00:00
ran:
  - .venv/bin/python -m pytest -q -n0 hunts/rh_moonshot/positivity_geometry hunts/rh_moonshot/probe_tower
  - .venv/bin/python hunts/rh_moonshot/positivity_geometry/challenge/circ_positivity.py
  - .venv/bin/python -m pytest -q -n0 hunts/rh_moonshot/positivity_geometry/challenge
outcome: no hidden RH or zero import; golden fake recomputed exactly to degree 300; the scope boundary over Q is conditional on GRH; the restricted-class label narrowed to a refutation of the function-field implication; Ihara zeta prior art on the phenomenon
artifacts:
  - hunts/rh_moonshot/positivity_geometry/challenge/circ_positivity.py
  - hunts/rh_moonshot/positivity_geometry/challenge/test_positivity_geometry_challenge_circularity.py
  - hunts/rh_moonshot/positivity_geometry/challenge/CHALLENGE_circularity.md

Compute: about 2 s of one core for the script, a few seconds for the tests; no estimate block needed (well under one core-hour).

Amendment 2026-10-26 (follow-up): field 5, prior art and published refutations

Added after the scorecard above was accepted; the original text is kept unchanged. All queries were web searches on 2026-10-26 (arXiv full text and author pages as the search engine indexes them; zbMATH Open, MathOverflow and the Lean Zulip archive targeted by query terms, not browsed directly). The hits are as the search results and abstract pages returned them; no paper below was read beyond its abstract or excerpts.

What changes.

• The window-2g lemma (NOTES 2.4) has a classical neighbour. The Caratheodory-Fejer theorem (hit 7, and the classical theory it cites) says that a positive semidefinite Hermitian Toeplitz matrix T_m of rank r < m is the moment matrix of r point masses on the unit circle. For the curve's multiset of 2g normalized eigenvalues, T_{2g+1} has rank <= 2g, so the theorem applies to the window of length 2g + 1. It does not by itself place the curve's own u_i on the circle: that needs an identification of the circle atoms with the multiset (weights and power sums), which this session did not derive. The NOTES' interpolation proof does the whole job directly at window 2g and is one step sharper. The builder's "probably known in the language of positive Toeplitz forms" (NOTES line 196) is therefore plausible but not confirmed by these searches; novelty stays "not established", and the lemma is recorded as adjacent to Caratheodory-Fejer.
• The golden fake's phenomenon has two published neighbours. Ihara zeta functions of non-Ramanujan regular graphs (hit 1) and Weng's original non-abelian zeta functions (hit 4) both have rationality and a functional equation and fail RH; the Ihara ones also have an Euler product with nonnegative prime counts. Neither has the curve's pole structure together with a +-1 local factor per place of F_q(t), so neither is the exact form. D3 above stands, now with citations.
• H_Q as a reformulation is in print (hit 2 and the Connes, Consani and Marcolli reference of field 5); no published proof or refutation of a Hodge index theorem for a square of Spec Z was found.
• Disposition: unchanged (unresolved; the golden fake a refutation of the function-field implication, D2). No hit refutes or proves the mechanism's step G3.

Correction to the run manifest above. Its finished time, 13:20, was written before the work finished and is not a measured time, nor is its started time; the follow-up manifest below has its finished time read from the clock and an approximate started time.

id: rh_moonshot-positivity_geometry-2026-10-26-challenge-circularity-field5
hunt: rh_moonshot
started: 2026-10-26T13:30+00:00
finished: 2026-10-26T13:38+00:00
ran:
  - web searches 1-7 of the table above
  - .venv/bin/python -m pytest -q -n0 hunts/rh_moonshot/positivity_geometry/challenge
outcome: Caratheodory-Fejer is a classical neighbour of the window-2g lemma, not a proof of it; Ihara and Weng zeta functions are published neighbours of the golden fake, not its exact form; disposition unchanged
artifacts:
  - hunts/rh_moonshot/positivity_geometry/challenge/CHALLENGE_circularity.md