We stand with people living with cystic fibrosis, their families, care teams, and researchers. This independent lab contributes donated time, research notes, and software experiments, not clinical care or medical advice. Support the wider work through the Cystic Fibrosis Foundation.

Donate
Sensei Ewok Research Lab Cystic Fibrosis AI Lab

Frontier Mathematics · Multi-Model AI Councils · Lean 4 & Scite Verification

Frontier Mathematics, AI Councils & Formal Proofs

An open, free-form observatory for mathematicians, computer scientists, physicists, and clinical researchers to explore machine-checked Lean 4 solutions, multi-model Frontier AI Councils, Scite citation polarity, and how formal verification grounds science, medicine, and technology. Independent open-source research, not medical advice.

Explore Top-Level Frontier Math & 3D Proof Models Showcase (/math/) → · ← Return to the AI-Loop Field Guide (/ai-loop/) · Home Page Math Showcase · Jump to Frontier Councils & STEM Grounding ↓ · Jump to 4-Proof Kernel & Scite Matrix ↓

Free-Form Three-Lens Explorer · 4 Solved Lean 4 Repositories · WebGL2 3D Proof Manifold

Interactive Lean 4 Proof Manifold & Three-Lens Scientific Observatory

In plain words: A machine-checked proof in Lean 4 verifies that every deduction from axioms to final theorem passes a tiny, unbribeable kernel check (0 sorryAx). Here you can explore all four landmark Lean 4 formalizations on their own mathematical terms first (Lens 1: Pure Mathematics & Theoretical CS), switch to Lens 2 (Frontier AI Councils, Scite & Science/Medicine/Technology) to see how multi-model falsification loops and empirical literature polarity work, or switch to Lens 3 (Two-Way CF Biophysics Bridge) to see how the exact same bottleneck and conservation laws govern protein folding (PDB 6MSM, 5UAK, 8EIQ) and genetic medicine. Research simulation only; not medical advice.

WebGL2 (#version 300 es · std140 UBO · 48 Instanced Wavepackets) · Drag or Arrow keys to orbit 3D Proof Manifold · Press R to reset
29,500 / 29,500 Kernel-verified Lean 4 theorems
75.31% Alluvial hypothesis flux (≥10% WT target)
3.76 kcal/mol Rate-limiting transition barrier (ΔG‡)
Wild-Type CFTR / Rescued 8EIQ Parallel CFTR biophysical state
Active Mathematical Axiom Perturbation None (prime exponent p ≥ 3, semistable Frey curve, Taylor-Wiles p=3/p=5 switch intact)
Emergent Counterexample or Kernel Verdict None — dim S_2(Gamma_0(2)) = 0 holds across all 29,500 Lean 4 theorems
Parallel Cystic Fibrosis Allosteric Mechanism Intact NBD1 thermal stability (Tm = 49.5 C) and cooperative NBD1-NBD2 ATP gating (Po = 0.34, 75.3% WT Cl- flux).
Selected Stage & Biophysical Focus (Click Any Stage Node or Row) 4. R = T Patching · FLT.DeformationRing.TaylorWilesPatching (fan-in 260 lemmas, depth 8.5, Delta G‡ = 3.59 kcal/mol) <-> Cooperative NBD1-NBD2 ATP dimerization & TM8 potentiation (PDB 6MSM / 8EIQ) [PDB 6MSM, 8EIQ · DOI: 10.1126/science.aaw7611]
Noether-Kirchhoff Deductive Flux Conservation CONSERVED: J_0 (100.00%) = J_5 (75.31% QED) + Sum D_k (24.69% dissipated) · Residual < 10^-9
3D Continuum Percolation & CFTR Allosteric Network (φ_c = 0.2896, p_c = 0.2488) SUPERCRITICAL: 65 nm PEG-LNP in 100 nm CF sputum (φ = 0.659 > φ_c, P_∞ = 0.761) · ETI 8EIQ bond network (p_bond = 0.68 > p_c)
Hilbert State-Vector Alignment (cos θ = ⟨ψ_k | e_QED⟩ & Orthogonal Projection) HILBERT ALIGNED: Stage 4 cos θ = 0.7819 (θ = 38.57°, bridge Δθ = −23.22°) · 5D Norm ||V|| = 1.9266 / 2.2361 · cos²θ + sin²θ = 1.0000
Kramers Thermal Escape Rate & Stokes-Brinkman Mucus Permeability (T-0485 / T-0486) ACCELERATED ESCAPE (6631.36x): k_Kramers = 0.29 -> 1940.29 s^-1 (MFPT 3417.7 -> 0.52 ms) · Stokes-Brinkman 2% ASL kappa = 459.5 nm^2 (l_B = 21.44 nm) vs 8% CF sputum kappa = 64.1 nm^2 (7.17x collapse)
Fiedler Algebraic Connectivity (λ2) & Cheeger Bottleneck Cut (T-0487 / T-0488) CHEEGER INEQUALITY VERIFIED (0.5*h^2 <= lambda_2 <= 2h): Stage 3 -> 4 (Ribet N->2 to Taylor-Wiles R=T) h(G) = 0.091 -> 0.486, Fiedler lambda_2 = 0.0441 -> 0.3049 (6.91x gain) · CFTR 8EIQ vs 5UAK (F508-R1070 & R347-D924): lambda_2 = 0.0404 -> 0.3261 (8.07x)
Active Explorer Lens (Pure Math · Frontier Council & STEM · CF Biophysics) Pure Mathematics & Theoretical CS Lens (Free-Form Proof Spine): Explore each proof on its own mathematical terms without requiring a biological mapping. Wiles (1995) and Taylor-Wiles (1995) prove that every semistable elliptic curve over Q is modular by patching deformation rings R_Q to Hecke algebras T_Q across auxiliary prime sets Q, forcing the Frey curve attached to a^p + b^p = c^p into the empty weight-2 level-2 cusp space S_2(Gamma_0(2)) = {0}.
Scite Citation Polarity vs. Lean 4 Kernel Parity (Observed Tallies · 2026-10-28) lean4export + comparator + nanoda (0 sorryAx, 29,511 repo thms / 29,500 DAG thms) · Math Anchor DOI 10.2307/2118559 (supporting=8, mentioning=1292, contrasting=0) · Biophysics/Clinical Anchor DOI 10.1126/science.ade2216 (supporting=126, mentioning=2120, contrasting=5, editorial_notices=0) · Stage 4 Provenance: [PDB 6MSM, 8EIQ · DOI: 10.1126/science.aaw7611] · Contrasting Calibration: All contrasting statements correspond to non-F508del Class I alleles or >200 nm unshielded particles above phi_c = 0.2896 [observed: Scite ML citation classification, accessed 2026-10-28]
Shannon/MDL Factoring Toy Case & AAV/mRNA Cargo Packing (T-0489 / T-0490) MDL WORKED EXAMPLE (computed toy case, not measured from any proof corpus): replacing 50 inline copies (16,000 bits) with one shared macro costs 820 bits (change −15,180) · AAV 4.7 kb Capsid: WT CFTR + flanking = 5,130 nt (+430 nt oversize · SEVERE TRUNCATION) vs ΔR minigene 4,626 nt (98.43% capacity · FEASIBLE · 74 nt headroom)
Jarzynski Non-Equilibrium Dissipation & CFTR ATP-Hydrolysis Power-Stroke Prover (T-0491 / T-0492) JARZYNSKI EQUALITY VERIFIED (⟨exp(−βW)⟩ = exp(−βΔF)): Dissipated work 4.61 -> 0.68 kcal/mol (85.32% cut, recovery err 0.0207 kcal/mol, ΔF = 3.76 kcal/mol) · CFTR ATP Power Stroke: WT η = 65% (W_mech = 7.93 kcal/mol, P_o = 0.42) vs F508del futile dissipation 10.37 kcal/mol (η = 15%)
Onsager-Machlup Action & CFTR Transition Path Sampling Prover (T-0493 / T-0494) LEAST-DISSIPATION GEODESIC VERIFIED: S_OM = 8.08 -> 2.90 (2.78x action gain) · CFTR Channel Gating Instanton: S_OM = 113.31 (5UAK) -> 11.53 (8EIQ, 89.82% action reduction) · Gating transition acceleration > 10^17-fold (τ_open = 1.8 ms)
Stochastic Thermodynamics & NESS CFTR Ion Efflux Prover (T-0495 / T-0496) TOTAL ENTROPY PRODUCTION VERIFIED (ΔS_tot = ΔS_sys + ΔS_env ≥ 0): Steady-state entropy generation rate 4.65 -> 1.08 k_B/step (76.77% reduction, steady proof flux J_proof = 94.20%) · CFTR Epithelial NESS Anion Efflux: WT J_Cl = 18.50 μA/cm², J_HCO3 = 3.80 μA/cm² (ASL depth 7.2 μm, luminal pH 7.35) vs F508del transport arrest J_Cl = 0.80 μA/cm² (ASL collapse 2.1 μm, pH 6.42)
Information Thermodynamics, Landauer Bound & CFTR Allosteric Feedback Prover (T-0500 / T-0501) LANDAUER DISSIPATION BOUND & SAGAWA-UEDA FEEDBACK VERIFIED (Q_diss ≥ k_B T ln 2 · I): Deductive branch mutual information I = 11.20 bits reduces proof erasure dissipation from 6.12 to 1.49 kcal/mol (75.7% reduction, feedback parameter γ = 1.36e-5) · CFTR Allosteric Mutual Information Feedback: WT I(NBD; TMD) = 0.88 bits, coupling α = 8.40, Po = 0.42 (Q_gate = 0.36 kcal/mol, η_allosteric = 88.0%) vs F508del decoupled gating arrest I(NBD; TMD) = 0.11 bits, α = 1.05, Po = 0.018 (futile dissipation Q_diss = 2.45 kcal/mol)
Five-stage formal proof to cystic fibrosis biophysical isomorphism
Stage & Theorems Lean 4 Module Scope Activation Barrier (ΔG‡) Alluvial Flux Pure Mathematical Mechanism / Parallel Domain Analog
1. Frey Curve (3,100)FLT.FreyCurve.SemistableReduction2.09 kcal/mol (VERIFIED)96.4%Primary CFTR transcript & nascent chain (Exons 1-27, 1,480 aa) [PDB 5UAK · DOI: 10.1038/ng.2745]
2. Mod-p Galois (4,900)FLT.GaloisRepresentation.TorsionPoints2.99 kcal/mol (VERIFIED)92.8%Co-translational TMD1/NBD1 assembly & ER chaperone folding check [PDB 5UAK · DOI: 10.1126/science.ade2216]
3. Ribet Descent (6,800)FLT.LevelLowering.RibetDescent3.76 kcal/mol (VERIFIED)86.1%ICL4-NBD1 interdomain docking & corrector pocket stabilization (VX-661/VX-445) [PDB 8EIQ · DOI: 10.1056/NEJMoa1908639]
4. R = T Patching (11,900)FLT.DeformationRing.TaylorWilesPatching3.59 kcal/mol (VERIFIED)79.4%Cooperative NBD1-NBD2 ATP dimerization & TM8 potentiation (PDB 6MSM / 8EIQ) [PDB 6MSM, 8EIQ · DOI: 10.1126/science.aaw7611]
5. S_2(Gamma_0(2)) = 0 (2,800)FLT.MainTheorem.CuspFormVanishing0.25 kcal/mol (VERIFIED)75.3%Open Cl- conducting pore state (8.4 pS, restored ASL ≥ 7.0 um) [PDB 6MSM · DOI: 10.1085/jgp.202012676]

Methodology · Multi-Provider AI Councils + Deterministic Kernels + Scite Polarity

How Frontier AI Councils & Formal Loops Ground Science, Medicine & Technology

Why agreement alone is never proof: Large language models can propose creative proof decompositions, algorithm reductions, and biophysical hypotheses, yet single-model generation is vulnerable to subtle hallucinations or unstated assumptions. Our architecture pairs diverse Frontier AI Councils with unbribeable deterministic verifiers so every surviving claim is backed by both syntactic proof and empirical provenance.

  • 1. Pure Mathematics & Theoretical Computer Science (Lean 4 Kernel Parity)

    Every mathematical benchmark in this observatory is exported via lean4export and audited by independent external typecheckers (comparator and nanoda) with zero sorryAx axioms. Whether patching deformation rings RQ ≃ TQ across 29,500 theorems in Fermat's Last Theorem (Wiles 1995), bounding critical-line simple zeros to 86.864% in XiPrime.lean (Ingham 1940), compiling sub-cubic 3SUM-to-APSP reductions (Williams & Williams 2018), or contracting 3-SAT tensor networks via SWAP = I − 2·singlet (Impagliazzo & Paturi 2001), any missing lemma immediately blocks compilation.

  • 2. Orthogonal Multi-Model Frontier Councils (Adversarial Falsification)

    Instead of asking models to vote, we assign orthogonal verification seats across distinct model architectures—including Claude Opus 5.5 for constitutional and mathematical rigor, GPT-OSS 120B for systems and formal boundary invariants, Gemini 4 Argon for cross-repository synthesis, and local GPU workers (Qwen 3.8 27B and Gemma 4 31B) for staged-diff security. A single reproducible counterexample from any seat blocks promotion until verified by a deterministic test.

  • 3. Scite Citation Polarity & Empirical Boundary Calibration

    In physics, medicine, and engineering, a deductive proof is only as valid as its physical boundary assumptions. We integrate Scite machine-learning citation classification (Nicholson et al. 2021) alongside our 113-source catalog to track supporting, mentioning, and contrasting citation statements across foundational DOIs. Contrasting citations pinpoint exact physical regime boundaries, such as worst-case unstructured k-SAT versus structured tensor-rank collapse or Class I nonsense CFTR alleles that lack protein substrate for small-molecule correctors.

  • 4. Two-Way Isomorphism Across Physics, Genetic Delivery & Clinical Medicine

    The same mathematical laws that govern proof DAG bottlenecks also govern biomolecular machines: Boltzmann transition-state barriers (ΔG‡), Noether-Kirchhoff flux conservation, Fiedler spectral connectivity (λ2) and Cheeger bottleneck cuts (PDB 8EIQ), Shannon/MDL cargo packing inside 4.7 kb AAV capsids (DOI: 10.1038/s41587-024-02194-3), Jarzynski non-equilibrium work (Jarzynski 1997), and Onsager-Machlup least-dissipation paths (Onsager & Machlup 1953). Explore the clinical and biophysical side of these models on the 3D CFTR workbench, Class I nonsense readthrough simulator, and in silico airway delivery explorer.

T2 Receipt · Lean 4 Comparator Audit, Scite Citation Ledger & Cross-Repo Companion Proofs

Companion Python Verification Engine: Every numerical invariant rendered on this page is cross-checked by 17 unit tests in cf-research/tools/discovery/formal_proof_thermodynamic_landscape.py and 94 deterministic falsification checks in discovery_council.py --falsify-all.

Evidence Catalog Lock SHA-256: 63eab6bb789bfb8096c6bc456e8809e6c79c7010ab83439cc933c77505a02cde (data/evidence.lock.json) · Independent open-source research, not medical advice.

Comparative Benchmark Reference · 4 Machine-Checked Lean 4 Formalizations

Kernel Audit, Bottleneck Invariants & Scite Polarity Matrix

Comparison of the four solved Lean 4 mathematical benchmarks, bottleneck mechanisms, Shannon/MDL compression ratios, and Scite citation tallies
Benchmark & Lean 4 Repository Verified Theorems Rate-Limiting Bottleneck & Bridge Mechanism Onsager-Machlup Action & NESS Entropy (T-0493 / T-0495) Foundational DOI & Scite Tally Interactive Manifold Link
A. Fermat's Last Theorem (anthropics/fermats-last-theorem) 29,511 repo / 29,500 DAG (0 sorryAx) Stage 4 Taylor-Wiles RQ ≃ TQ patching (11,900 thms, 260-lemma fan-in, 8.39 → 3.59 kcal/mol via p=3 ↔ p=5 switch) SOM = 8.08 → 2.90 (2.78×); ΔStot = 4.65 → 1.08 kB/step (−76.77%) DOI: 10.2307/2118559 (supporting=8, mentioning=1292, contrasting=0) Load FLT in 3D Manifold ↑
B. Riemann Zeta & XiPrime (anthropics/formal-math/zeta23) 1,420 Lean 4 thms (0 sorryAx) 86.864% (≥ 0.86864) simple critical-line zeros via stationary-phase mollifier (MK = 4.6248, 7.88 → 3.48 kcal/mol) SOM = 6.70 → 2.53 (2.65×); ΔStot = 4.38 → 1.02 kB/step (−76.71%) DOI: 10.1515/crll.1989.399.1 (supporting=1, mentioning=12, contrasting=0) Load Zeta/XiPrime in 3D Manifold ↑
C. Sub-Cubic 3SUM → APSP (anthropics/formal-math/3sum-apsp) 980 Lean 4 thms (0 sorryAx) Williams-Williams O(n1.9992) 3SUM → O(n2.99942) APSP + Fredman 64× decision-tree bit-packing (7.66 → 3.46 kcal/mol) SOM = 6.36 → 2.35 (2.71×); ΔStot = 4.25 → 0.98 kB/step (−76.91%) DOI: 10.1137/15M1024524 (supporting=0, mentioning=100, contrasting=0) Load 3SUM-APSP in 3D Manifold ↑
D. ETH Refutation: 2o(n) #3-SAT (anthropics/eth-3sat) 493 PaperIndex thms (0 sorryAx) theorem ETHIsFalse via SWAP = I − 2·singlet, S2_tilde_rank_three (rank 3 < 4, Δlog2 = 0.41504 bits/triplet, 8.10 → 3.60 kcal/mol) SOM = 7.42 → 2.82 (2.63×); ΔStot = 4.48 → 1.04 kB/step (−76.74%) DOI: 10.1006/jcss.2000.1727 (supporting=1, mentioning=880, contrasting=0) Load ETH #3-SAT in 3D Manifold ↑