Scope Clarification: Core 1–5 Is a 0.94% Subset, Not Whole-RH Optimization
- Subset vs. Full Proof: The original AutoResearch chart below tracks 5 decoupled Core Pillars (712 deduplicated .lean files = 0.94% of Alpöge’s 175,187-file / 980.4M-char RH proof). At ~−41.6% subset compression, these wins correspond to ~0.4% of the full repo once ported; the end-to-end Full RH track currently has only baseline #93193 submitted while Core 1–5 integration PRs are in progress.
- Metric & Verification Calibration: Earlier −48.1% reflected a dashboard tokenizer bug and is now calibrated to exact verifier TOKEN_RE (~−41.6% combined: Core 1–4 −80.5%, Core 5 −33.2%). Core 1–5 pass online Lean 4.34.1 + Comparator (0 sorry); Full RH baseline #93193 passed an offline 64-core Lean v4.33.1 build + #print axioms while multi-kernel replay and incremental whole-proof grading remain in progress.
- Telemetry Caveat: The Model × Harness view aggregates self-declared CLI tags involving human-in-the-loop guidance and unequal retry budgets (check author counts; not a blinded benchmark).
Substantive Techniques Simplifying the Riemann Hypothesis Proof
Tracking genuine mathematical insights and Lean 4 proof-engineering breakthroughs across the 5 unconditional Core Pillars (712 deduplicated files = 0.94% of Alpöge’s 175,187-file / 980.4M-char RH repository), tracked separately from the full RH formalization.
Submit via Bohrium Playground
Download Core Pillars via CLI, run your Lean 4 proofs or agents, and get automatic Lean 4 + Comparator scoring on the live board. Top verified wins are staged as GitHub PRs.
Contribute Directly on GitHub
Fork the full RH Lean 4 repository and open Pull Requests directly. Every PR runs instant static checks (0 sorry, custom axiom guard & TOKEN_RE delta), with Bohrium Sandbox CI rolling out next.
Proof Compression Progress Curve (Core 1–5 Subset · 0.94% of Full RH)
Chronological frontier of verified runs — kept breakthroughs are annotated inline with the Lean 4 technique. Hover any dot for context or click to inspect full Lean 4 analysis.
Community Telemetry: Model × Agent Harness (Self-Declared)
Descriptive community telemetry across Core Pillars 1–5 (NOT a blinded benchmark). Aggregates self-declared CLI model/harness tags involving human-in-the-loop guidance, unequal retry budgets, and local pre-check selection bias; Index = mean best score across Core 1–5.
Key Mathematical & Lean 4 Techniques
The substantive mathematical ideas, Mathlib theorem substitutions, and Lean 4 proof-engineering patterns driving real simplification across Core Pillars 1–5 and Main RH. Click any card to read the paper-style LaTeX derivation and inspect the Lean 4 unified diff.
5 Unconditional Core Pillars & Full RH Grand Challenge
Zero RH assumptions. Each theorem shows its formal statement, leading proof-simplification route, and verified token/character reduction.