Project
Recursive Self Improvement Lab
A verifier-gated research program assembled from sealed local experiments: propose, intervene, verify, archive, and inherit only when held-out evidence improves.
Status Active
Recursive Self Improvement Lab
An attempt to build a recursive self-improvement lab from a portfolio of sealed research projects rather than from a single self-modifying agent. Improvement counts only when an external truth standard — exact oracles, Lean gates, sealed artifact hashes — accepts a claim the proposer cannot rewrite, and when archive-conditioned reuse later beats memoryless baselines on held-out families.
Club A is the evidence surface: Wave-Causal Circuit Search, Parameter Golf, Gigatoken Lab, and Nostradamous — each with researcher-readable reports and interactive or snapshot notebooks. Club B is the substrate: Open Problems Lab and Lorenz for formal verification; Sansara and Geist for agent operation; Vault Ops for archive and publication; Agora and AlphaEvolve for simulation and evolutionary scaffolds.
This page is the program map. Individual claims live on the project and report pages; negative results stay published beside positive ones. The first inheritance experiment remains queued behind Wave-Causal Tier 0 semantics and is not advertised as running recursion.
Reports
- Report
Cognitive Systems Portfolio Thesis
How Memory Store (cognitive memory at Julep), phi9, Sansara, Agora agent simulation, and the RSI Lab form one research program: durable world models, verifier-gated discovery, and agents that do not own the control plane.
- Report
Recursive Self Improvement Lab — Portfolio Map
How Club A evidence engines and Club B method infrastructure connect into a verifier-gated recursive discovery program — and what is still not claimed.