Autonomous Scientific Discovery
solver.press is the public window into AegisMind's closed-loop discovery engine. Hypotheses are generated, stress-tested through five-model adversarial debate, formally verified, experimentally validated on Google TPUs, and published — without human bottlenecks.
Four peer preprints across drug discovery, ML theory, and quantum-hardened encryption published in 17 days.
Highest-confidence discovery
Computer Science · Quantum annealing-based subgraph isomorphism algorithms can identify structural motifs in protein-ligand docking data th…
Read full discovery →80+ discoveries generated. 4 published. Loop running since March 2026.
Source: AegisMind Research
Read full discoverySource: AegisMind Research
Read full discoverySource: AegisMind Research
Read full discoverySource: AegisMind Research
Read full discoverySource: AegisMind Research
Read full discoverySource: AegisMind Research
Read full discoveryIngest
arXiv and Semantic Scholar papers across 15+ scientific domains are continuously embedded into a vector store and searched for cross-domain bridges invisible to single-domain researchers.
Generate & Debate
Five frontier AI models (Claude, GPT-4, Gemini, Grok, Mistral) independently generate hypotheses then critique each other in adversarial debate. Hypotheses that survive critique are scored. Only high-confidence survivors are published.
Formal Verification
Z3 checks logical consistency of surviving hypotheses — not empirical truth, but internal coherence. Unverified hypotheses are flagged rather than suppressed.
Experimental Validation Package
High-confidence discoveries receive a full EVP: precise quantitative hypothesis, disproof criteria, experimental protocol, abort checkpoints, implementation code, GPU hour estimates, cost range, and ROI projection. Ready to execute.
Execute, Close the Loop
OceanSparx runs EVPs on Google TPU Research Cloud. Results validate or refute — closing the loop or opening new downstream discovery avenues. Validated findings are published as citable peer preprints with DOIs.
Hypothesis Aggregation Papers
View all →Formal papers synthesising solver.press discoveries into testable hypothesis clusters with complete experimental validation packages.
Graph-theoretic hypothesis: SCCs of size ≥3 in the directed collateral sensitivity graph define closed evolutionary traps. Z3-verified. If confirmed: first mathematically closed antibiotic cycling framework, ~50,000–200,000 deaths averted/year.
Combined QS-inhibitor plus QS-dependent antibiotic therapy creates doubly unfavorable evolutionary landscape for resistant mutants. Lotka-Volterra public-goods model predicts selection coefficient s ≤ −0.05 under combined therapy.
Computationally validates exact O(ε) convergence of performatively stable solutions to classical SP optima across 5 problem families. α = 1.000–1.028, R² ≥ 0.9995. Proportionality constant explicitly characterized: C = L_D·‖x*(0)‖·(1+O(ε)).
Nash equilibrium cavity detuning achieves 84.9% ergotropy improvement over resonance (p < 10⁻³⁵). Loewner matrix interpolation identifies optimal coupling g* = 0.01 with 54.7% improvement using 13 nodes, 0% prediction error.
Published Research
Four peer preprints generated by the AegisMind discovery loop.
Three-method convergence (virtual screening, Nash equilibrium optimisation, Bayesian optimisation) across 2,639 FDA-approved compounds. Eptifibatide identified as three-way convergent hit. CNS-penetrant analogues Ponatinib and Entrectinib identified.
Loss landscape topology across number formats and multi-target drug discovery. Scaling law (FP32↔BF16 barrier ∝ params⁻⁰·⁸⁵, R²=0.98). Multi-target Bayesian optimisation across six therapeutic targets including KPC-3 (AMR) and MSH3 (HD).
Four-phase pipeline across 36,966-cell scVI atlas identifying ZNF740/BRD3, Cathepsin S, and DNMT1 as novel therapeutic axes in CA-RIM lesions. No approved disease-modifying therapy exists.
Quantum-hardened encryption grounded in relativistic causality. 17,255 bits quantum entropy from 30-minute NICER observation. Immune to Shor's and Grover's algorithms by design. Three Australian provisional patents filed.
solver.press publishes a curated selection of AegisMind's discoveries. Research teams, pharma BD teams, and technology organisations can engage with the full discovery engine at aegismind.app — querying specific domains, commissioning targeted discovery runs, and receiving complete Experimental Validation Packages ready for laboratory execution.
Each EVP includes: precise quantitative hypothesis · disproof criteria · full experimental protocol · abort checkpoints · implementation code · GPU hour and cost estimates · ROI projection · prerequisite dependency map · downstream discovery unlocks.
The engine is running. The loop is closed.
Access the discovery engine at aegismind.app →