1.18.0 — Meta-Agentic AGI v3 Curriculum Lab

The identity-only entry point now has a substantive companion: a reproducible curriculum lab that generates integer-program tasks, independently infers hypotheses from examples, evolves solver configurations and tests the frozen winner on a separate stream. Actual correctness and operation counts drive Pareto selection; explicit adaptation and a cost/entropy proxy make the research loop inspectable.

Four starter experiments cover balanced skills, missing curriculum coverage, resource limits and deeper composition. The responsive browser interface exposes training rounds, examples, inferred programs, candidate metrics, parent/child lineage, independent gates, explicit settings recovery and six-file verified exports. The standard-library CLI, packaged loopback server and executable notebook provide the same workflow locally.

Legacy generated-code paths now share Docker isolation and cannot fall back to host execution. The provider curriculum asks an independent solver without revealing the answer. Rate-limiter and retry behavior are bounded, free-energy inputs are validated, and the royalty example fixes aggregation and currency accounting while refusing unsupported payments. Every original diagram, research narrative, notebook and visual replay is retained.

Publication requires native/browser parity, boundary and accounting regressions, installed-wheel acceptance, real browser journeys, mobile/accessibility checks and exact-commit public-site evidence. The lab is finite synthetic program induction; it neither trains model weights nor authorizes enterprise deployment.

Use Start here for installation and the Curriculum Lab guide for experiments and recovery.