Zeta Lab Research Portfolio Recommendation

Transcribed from Zeta Branches - Hunt.pdf (19 pages, dated August 11, 2026; audience "Zeta Lab research leadership / research-allocation decision maker"). Mathematical typography restored where PDF extraction garbled it; content otherwise unaltered. This is an outside recommendation, not a result — see docs/reviews/README.md for the directory contract and ROADMAP.md ("The outside memos, triaged") for what was adopted from it.

Purpose: Identify the highest-value live research fronts in the current Zeta Lab repository and recommend how available parallel-agent capacity should be allocated.

Executive Summary

Zeta Lab is currently in an unusually favorable position.

Several exploratory programs have already eliminated their cheapest or most misleading formulations. That matters because the surviving problems are no longer vague searches for "something that might work." In multiple areas, the repository has isolated sharp mathematical frontiers with:

• known failed relaxations;
• explicit remaining gaps;
• reusable computational infrastructure;
• adversarial regression tests;
• quantitative targets;
• identifiable next lemmas.

The recommendation is therefore not to concentrate all capacity on a single continuation of the current best bound. Instead, Zeta Lab should operate a parallel research portfolio centered on three major programmes:

• Flagship — Full-data zero-configuration realizability / Level-6 counting dual. This is the most compelling immediate frontier. The existing scalar relaxation collapses to the known 0.6725007 result, but the richer ordered-configuration problem does not. The remaining gap toward the approximately 0.68185 configuration ceiling represents information that has genuinely not yet been extracted.
• Theory Builder — General higher-ξ derivative hierarchy. The correction of the ξ″ coefficient/form-factor machinery should be generalized from one repaired case into a theory of ξ⁽ᵏ⁾/ξ⁽ᵏ⁺¹⁾, C_{k,i}, F_k(α) for arbitrary derivative order k.
• Moonshot — Global positive structure assembled from prime-local arithmetic. The naive place-by-place positivity programme has failed, but that failure sharply identifies a deeper question: whether local arithmetic objects can be coupled globally into a canonical positive pairing reproducing the Weil criterion.

These programmes are mathematically distinct enough to run concurrently, while sharing Zeta Lab's verification and adversarial infrastructure.

1. Highest Priority: Full-Data Configuration Realizability

Why this deserves the largest allocation

The central numerical interval — 0.6725007 to approximately 0.68185 — should not be viewed as a small optimization gap.

The repository has already established that the obvious scalar/pair-measure version of the problem collapses back to the existing single-window bound. That negative result is important: it identifies exactly what information is missing.

The missing information lives at the level of actual zero configurations:

• ordering;
• multiplicities or marks;
• neighboring-gap structure;
• overlap between pair constraints;
• consistency between local configurations;
• realizability of candidate pair distributions by a genuine ordered point process.

The richer configuration problem therefore remains substantially different from the dead scalar LP.

Recent work further narrows the target to a mixed-depth two-dimensional counting problem involving the interaction kernel T(dt, y, y′).

The existing level-4 machinery reportedly exhibits a measurable budget deficit rather than an indefinite qualitative failure. Two different sharpenings appear large enough, at least numerically, to potentially close that deficit.

That combination is rare:

a known theorem barrier + a quantified deficit + multiple plausible sources of recoverable slack.

Recommended subteams

Allocate 6–7 agents.

• Team A — Mixed-depth counting dual. Derive the strongest possible two-dimensional dual involving dt, y, y′. Retain depth information that previous scalar projections discarded. The immediate objective is a rigorous Level-6 inequality rather than another numerical relaxation.
• Team B — Recover discarded payments. Audit the level-4 counting argument for losses caused by: non-adjacent interactions; crude multiplicity caps; discarded overlapping constraints; uniform bounds replacing geometry-sensitive bounds. Measure the exact amount recoverable from each refinement.
• Team C — Collective charging / local potentials. The existing pointwise pair charging appears potentially much looser than necessary. Search for a local potential or collective energy argument in which several neighboring pairs pay jointly rather than independently. This is one of the most plausible mechanisms for beating the current deficit.
• Team D — Configuration adversary. This team should not construct proofs. Its task is to generate hostile realizable configurations: periodic lattices; perturbed lattices; clustered configurations; alternating-depth patterns; quasi-periodic configurations; configurations near the empirically dangerous spacing around 0.6; high-multiplicity or mixed-mark examples where permitted. Any proposed inequality must survive this team.
• Team E — Exact certification. Translate promising numerical duals into: rational certificates; interval enclosures; exact finite LP duals; formally checkable inequalities where practical. The numerical optimizer must never be the final authority.
• Team F — Independent theorem reconstruction. Re-derive the zero-side machinery directly from the pinned upstream definitions. This team must specifically guard against another transpose-versus-conjugate-transpose substitution.

Definition of success

A success is not necessarily reaching 0.68185.

A major success would be either θ_full > 0 for a genuinely configuration-sensitive correction that propagates through the exact zero-side theorem, or a rigorous dual obstruction proving that bandwidth-one information cannot improve the 0.6725007 result through this entire configuration class.

Either outcome materially changes the frontier.

2. Build the General Higher-ξ Hierarchy

The ξ″ investigation appears to have become much more important than its original objective. What began as a check of historical percentages has produced:

• corrected coefficient arithmetic;
• an explicit generating mechanism;
• a controlled coefficient tail;
• a rebuilt logarithmic-derivative representation;
• a new bridge argument extending beyond an apparent 1/2 barrier.

The next question should no longer be "What else can we calculate for ξ″?" It should be:

What is the general theory for zeros of derivatives of ξ?

For R_k(s) = ξ⁽ᵏ⁾(s)/ξ⁽ᵏ⁺¹⁾(s), derive a uniform mechanism for arbitrary k.

Research targets

Seek formulas or recurrences for C_{k,i} and the associated form factors F_k(α). Determine whether there are general laws governing:

• signs;
• rationality;
• generating functions;
• coefficient asymptotics;
• radius of convergence;
• entire continuation;
• behavior as k → ∞;
• dependence of the local zero process on derivative order k.

Use multiple independent constructions:

1. Bell-polynomial / differential-algebra derivation;
2. Dirichlet-convolution derivation;
3. symbolic generating-function derivation;
4. independent exact implementation.

Allocation

3–4 agents.

High-value outcome

A single general theorem generating the corrected higher-derivative arithmetic would transform the Bian correction from a historical repair into a reusable new theory.

3. Push the ξ″ Bridge from 0.51 Toward Bandwidth One

The current ξ″ work suggests a recurring pattern: several apparent bandwidth barriers have turned out to be artifacts of lossy estimates.

The most recent extension beyond 1/2 is especially significant because it arose from preserving the true logarithmic frequency separation instead of collapsing it into a cruder cutoff estimate.

The right next question is therefore:

What genuinely fails at 0.6, 0.75, 0.9, 1 − ε?

Do not merely rerun the same proof with larger parameters. Construct a complete exponent budget for every analytic error term as a function of α. At each target bandwidth, identify the first inequality that ceases to close. Then classify that obstruction as:

• fundamental analytic barrier;
• avoidable Cauchy–Schwarz loss;
• cutoff artifact;
• finite-prime problem;
• contour choice;
• overly crude norm;
• lack of cancellation.

Allocation

3 agents, plus an independent adversarial reviewer.

Best possible outcome

A theorem valid throughout every compact subset of |α| < 1. A rigorous proof of a genuine barrier below one would also be valuable.

4. Turn the Completed-CUE Oracle into Mathematics

The completed-CUE derivative experiments currently serve as a powerful independent numerical oracle. That should be upgraded into a theorem programme. The key question is:

F_k^CUE(α) =? F_k^arith(α)

after the appropriate scaling and normalization.

Instead of merely observing agreement between random matrices and arithmetic, derive the limiting derivative-zero process analytically.

Approaches to parallelize

• determinantal-process methods;
• characteristic-polynomial identities;
• asymptotic random-matrix analysis;
• differential identities for completed characteristic polynomials.

Allocation

2–3 agents.

Why this is unusually attractive

Success would create a three-way equivalence:

differential algebra ⟷ Dirichlet arithmetic ⟷ random matrices.

That would explain the higher-ξ coefficients structurally rather than merely reproduce them.

5. Determine the Full Local Process of Derivative Zeros

Pair correlation captures only second-order information. A more ambitious programme should ask what differentiation does to the entire local GUE/CUE-style zero process.

Targets include: nearest-neighbor spacing, 3-point correlation, n-point correlation, repulsion exponents, clustering statistics, and the dependence of these quantities on derivative order.

The conceptual question is:

Does differentiation induce a universal transformation on the local zero point process?

This could potentially unify several otherwise isolated observations in the higher-ξ work.

Allocation

2–3 agents. This should be regarded as a theory-building programme rather than an immediate bound-improvement project.

6. Identify the Analytic Function Behind Corrected F₂

The coefficient series for corrected F₂ appears computationally well controlled. That makes the next problem conceptual:

What function is it?

Search systematically for:

• differential equations;
• integral representations;
• functional equations;
• hypergeometric forms;
• Bessel-type representations;
• continued fractions;
• explicit transforms;
• coefficient asymptotics;
• singularity or entire-function structure.

Numerical recognition methods such as PSLQ may be used for discovery, but no recognized identity should be promoted without proof.

Allocation

2 agents. A successful closed or structural representation may simplify both bandwidth-one analysis and the general-k hierarchy.

7. Complete the Kernel-Checked Davenport–Heilbronn Result

This is one of the strongest near-term formalization opportunities.

The architecture already appears largely present for a machine-checked theorem showing that zeta-like functional symmetry alone cannot imply RH, using the Davenport–Heilbronn example.

The main remaining cost is certified evaluation.

A promising optimization is to replace rectangular complex enclosures for m^{−s} with a polar or mean-value representation that avoids significant dependency inflation.

Allocation

2 agents. One owns the analytic enclosure. One owns certificate generation, Lean integration, and final kernel checking.

Win condition

A zero-sorry, independently checkable theorem in the proof assistant. This would be a clean standalone contribution even if none of the numerical zeta bounds move.

8. Upstream Formal Mathematics into Mathlib

Two particularly good formalization targets have emerged.

• Sturm root counting. This is broadly useful outside the zeta project and appears to fill a genuine general-purpose gap.
• Hardy Z. A formal Hardy Z construction would provide the correct foundation for future critical-line computations and may avoid some of the branch-management difficulties associated with direct complex logarithms of Γ.

Allocation

2 agents, operating mostly independently. Avoid duplicating other ongoing formalization efforts where equivalent theorems are already being developed elsewhere.

9. Moonshot: Solve the Local-to-Global Positivity Problem

The earlier local-positivity experiment found that individual prime/place contributions can sometimes be represented as norms. However, the local pieces do not consistently possess the sign required to make naive prime-by-prime positivity work globally. That route is closed in its simple form.

The more interesting successor is:

What additional global structure would make the local arithmetic pieces assemble into a single canonical positive pairing?

Potential structures include:

• cross-place coupling;
• global cocycles;
• intersection pairings;
• cohomological constructions;
• operator models;
• constrained Hilbert-space embeddings.

The desired architecture is:

prime arithmetic → global object → canonical pairing → ‖Φ(f)‖² → Weil positivity.

Mandatory adversaries

Any construction must structurally exclude non-RH rivals such as Davenport–Heilbronn and inappropriate Epstein combinations. It is not sufficient for the proposed representation to "work numerically" for ζ.

Allocation

2 long-horizon agents. This is low-probability but potentially transformational.

10. Generalize the de Bruijn–Newman Repair Clock

Previous experiments suggest that certain collision times under heat flow may be determined largely by the geometry of the initial zero configuration rather than by arithmetic coefficients. That observation should be generalized across families.

Build a parametric heat-flow framework for completed L-type kernels and ask:

• which flow observables are universal consequences of zero geometry?
• which retain genuine arithmetic information?
• can zero-dynamics approximations be rigorously connected to the underlying PDE?

Allocation

2 agents. The desirable outcome is a theory separating universal zero dynamics from genuinely arithmetic flow invariants.

11. Revisit Euler-Product Defect vs. Zero Geometry — Correctly

The earlier attempt to correlate factorization defect with a zero-position statistic was not methodologically valid enough to support its conclusion.

Do not simply rerun the correlation. Invert the research question. Ask:

Are Euler-product structure and zero geometry fundamentally independent coordinates, or is there a deeper joint invariant connecting them?

Construct families in which factorization defect and zero geometry can be varied independently. Attempt either to:

• discover a real structural coupling; or
• prove an independence theorem/counterfamily showing no scalar relationship can suffice.

Allocation

1–2 agents. A rigorous negative theorem would be a successful outcome.

12. Build a Certified Adversarial Weil-Function Explorer

The existing Weil machinery can be turned into a much stronger experimental platform.

One agent searches large spaces of admissible spectral factors/test functions for extremal or suspicious behavior. A second independent backend certifies candidate values rigorously. A third implementation should eventually duplicate the certification without sharing the same upstream transformation code.

The purpose is not to infer RH from finite positivity testing. The purpose is to:

• stress-test structural conjectures;
• locate extremal test functions;
• generate counterexamples;
• identify where candidate positivity principles actually become sharp.

Allocation

2–3 agents.

13. Make Verifier Independence Measurable

One of Zeta Lab's most important methodological findings is that two independent numerical backends can agree and still be wrong when they share the same faulty upstream transformation.

Therefore:

"checked by two implementations" is not enough.

Build a provenance DAG for every important computation. Record shared ancestry in:

• source data;
• parsers;
• normalizations;
• mathematical transformations;
• discretizations;
• contour choices;
• intermediate formulas;
• certification policies.

Then quantify something like an independence radius:

At what earliest mathematical/implementation layer do the two verification paths become genuinely independent?

Allocation

2 infrastructure agents. This would improve the reliability of every other programme.

14. Run a Repository-Wide Guard Offensive

The existing verification work has already shown that guard failures are a meaningful source of false confidence.

A dedicated adversarial team should inspect every important guard and ask:

What exact incorrect computation is this guard supposed to detect, and has that detection power actually been demonstrated?

For each guard:

1. identify its intended lesion;
2. construct the smallest mutant exhibiting that lesion;
3. confirm the guard fires;
4. identify nearby lesions it does not detect;
5. document its true scope.

Allocation

2 destroyer agents. They should do no feature development.

15. Complete the External-Referee Experiment

Zeta Lab's verification methodology should be tested by outsiders — or by agents operating without access to the internal cultural assumptions that created it.

Give a clean specification to an isolated team and see whether it can independently construct a meaningful adversarial battery.

Likewise, stronger external semantics tools such as Alive2 should be incorporated where applicable in compiler-oriented experiments.

Allocation

1–2 agents. This tests whether Zeta Lab's methodology is genuinely transferable rather than internally self-consistent.

16. Turn the Equivalence Web into a Research Scheduler

The repository already records relationships among methods and statements. That information should become operational.

Construct a theorem/dependency graph linking areas such as: pair correlation; simplicity; zero density; derivative-zero statistics; explicit formulas; moments; Weil positivity; Li/Jensen-type criteria; Newman flow; Selberg-class structure.

Then rank unresolved statements by their downstream leverage. The scheduler should favor questions where one resolution:

• opens many paths;
• kills many redundant paths;
• resolves multiple equivalent formulations;
• invalidates a broad family of heuristics.

Allocation

1–2 agents. This could substantially improve the efficiency of future parallel-agent research.

Recommended Allocation for ~30 Research Agents

A reasonable first allocation is:

The numbers should remain flexible. Agents should be reassigned when a branch hits a proven barrier rather than being kept alive merely because capacity was originally assigned to it.

Organizational Rule: Constructors and Destroyers Must Be Separate

The strongest lesson from Zeta Lab's recent work is methodological.

Do not assign every agent to construct a proof. For major claims, maintain three roles:

• Constructor. Find the strongest possible theorem.
• Independent reconstructor. Derive the result by a genuinely different mathematical route.
• Destroyer. Assume the theorem is false. Search specifically for: normalization failures; transposes replaced by adjoints; shared implementation ancestry; hidden completeness assumptions; invalid limit interchange; non-realizable extremizers; missing tails; finite-grid artifacts; rival examples.

For important results, the destroyer should ideally work without seeing the constructor's detailed reasoning until it has independently generated hostile examples.

Explicit Do-Not-Fund List

Unless materially new information appears, avoid allocating research capacity to:

• another scalar/pair-measure LP;
• generic window reoptimization with unchanged information;
• the withdrawn transpose-to-adjoint positivity construction;
• attempts to bypass the λ > 1 obstruction without new prime-side input;
• another unchanged CGdL transplant;
• naive prime-by-prime positivity;
• the closed finite Poisson-cokernel matrix route;
• Lehmer heuristics based only on small |Z|;
• rediscovery of Bian's historical percentages;
• numerical optimization before the underlying arithmetic object is certified.

These mechanisms have already provided their useful information.

Recommended Executive Decision

Authorize three principal programmes immediately:

1. Flagship — Full-data configuration realizability / Level-6 mixed-depth dual. Give this the largest team and explicit authority to pursue either a positive correction beyond 0.6725007 or a rigorous impossibility result.
2. Theory Programme — General F_k hierarchy for zeros of derivatives of ξ. Treat the repaired ξ″ calculation as the k = 2 instance of a potentially general theory.
3. Moonshot — Global positive pairing from prime arithmetic. Pursue the deeper local-to-global structure exposed by the failure of naive placewise positivity.

At the same time, maintain an independent verification division covering:

• adversarial configurations;
• theorem reconstruction;
• guard testing;
• verifier-independence analysis;
• exact/interval certification.

Bottom Line

Zeta Lab should not behave as though it has one promising idea left.

It currently has multiple mathematically distinct live frontiers, several of which became clearer precisely because earlier approaches failed. The highest-value strategy is therefore a portfolio:

configuration-level zeta frontier + general derivative-zero theory + structural moonshots + hostile verification.

The full-data configuration problem deserves the strongest immediate push because it sits directly at a known quantitative frontier and retains information that all cheaper relaxations discard.

The higher-ξ programme deserves simultaneous investment because the correction of F₂ appears to have exposed a broader derivative-zero theory waiting to be developed.

The structural programmes should remain alive because a genuinely new route to RH is unlikely to look like another decimal-level optimization of machinery whose information ceiling is already understood.

The operating principle should be: exploit known sharp fronts aggressively, while maintaining enough independent bandwidth to discover an entirely different theorem.