Opened 2026-08-17 on branch claude/riemann-hypothesis-research-ofds8s.
This hunt is a research campaign, not a single conjecture. Its mandate: survey the current RH-adjacent frontier from primary sources, generate a large portfolio of concrete attack surfaces, and drive the most promising ones through a discovery, destruction, and rigor pipeline until one produces the largest defensible new mathematical statement this run can support. A closed route with a witness counts as output. A rediscovery does not.
The campaign deliberately does not privilege the laboratory's existing flagship (the critical-line proportion candidate in hunts/frontier_math/). That work is prior art for this run: anything this hunt produces must be new relative to it as well as to the literature.
id: rogue_frontier
question: Which RH-adjacent frontier admits the largest new mathematical advance this run can produce and defend, and what is that advance?
frontier: multiple fronts; each candidate project records its own frontier line with citations in FRONTIER_MAP.md before work starts
proposed_attack: portfolio search: parallel literature mapping, 30+ scored candidate projects, escalating tiers of computation and proof, adversarial destruction before any promotion
dead_routes:
- re-deriving the frontier_math retention chain (that is hunts/frontier_math/ prior art, not this hunt)
- direct numerical verification of RH to a new height as a headline (compute-bound, no new mathematics)
- heuristic GUE analogies promoted without a theorem or an enclosure
required_oracles:
- exact rational or integer arithmetic checked by an independent implementation
- ball or interval enclosures via zeta.rigor with both backends where feasible
- the Lean 4 kernel with zero sorrys and standard axioms
- mpmath reference routines (zetazero, siegelz, nzeros) as numerical cross-checks
- located primary literature quoted with theorem numbers
kill_conditions:
- a claimed improvement fails an independent recomputation
- the novelty search finds the statement in prior art
- the effect does not survive increased precision or an enclosure
- a structure-matched rival (Davenport-Heilbronn via zeta.epstein.battery) satisfies the same claimed criterion
- the derivation is found to secretly assume RH or an unproved hypothesis without saying so
agents_may:
- search
- derive
- code
- attack
- formalize
agents_may_not:
- declare novelty
- declare theorem status
- promote their own claimScope
May write: hunts/rogue_frontier/**, figures/ if a figure earns it, and at most one docs/NN-*.md at the end if a result earns one (next free number taken from the tree at write time, never guessed). May not touch zeta/, ontology/, harness/, or other hunts except to read them. The Lean arm under lean/ may gain new files only for a claim that has already survived the destruction stages, in its own module, without touching existing proofs.
Grading
Claims here use the repository ladder: measured, hardened, kernel-checked. The word owned by zeta/rigor.py is not used in this directory at all, per tests/test_hunt_probe_discipline.py. A composite claim takes the grade of its weakest step. Ledger files:
FRONTIER_MAP.mdstate of the art with primary citationsIDEA_PORTFOLIO.mdcandidate directions and scoresRESULTS_LEDGER.mdactive, dead, promoted, refuted claimsNOVELTY_LEDGER.mdsearches performed, closest prior artFAILURE_LEDGER.mddead ends and why they diedREPRODUCE.mdexact reproduction steps for anything promoted