A model-agnostic reasoning protocol for difficult, ambiguous, and multi-step problems. Instead of committing early to one answer, it maintains materially different candidate paths, verifies them independently with evidence and tools, detects shared assumptions and contradictions, reallocates compute as uncertainty changes, preserves recoverable dormant branches, and collapses only when one result is sufficiently supported. It runs on classical AI models and hardware; it is quantum-inspired reasoning, not quantum computation.

multi-branch reasoning · falsification · adaptive compute · branch revival · model-agnostic skill

An independent deterministic verification supervisor for AI coding agents. It turns requirements and implementation plans into auditable execution contracts, tracks dependency/output DAGs, can automatically compile structured plans into conservative STRIPS contracts, verifies reachability and exact recompilation, seals verification state cryptographically, records hash-chained execution evidence, and gates completion across single- or multi-agent work with PASS / FAIL / UNKNOWN.

v2.4.0 · agent verification · STRIPS/PDDL · cryptographic sealing · evidence chains

A versioned scientific modeling engine and Agent Skill built around a machine-readable Model IR. It supports native algebraic, ODE, PDE, DAE, stochastic, optimization, control, network, Bayesian, agent-based, discrete-event, hybrid, multiphysics, and causal modeling workflows, with fitting, uncertainty analysis, causal identification, Bayesian diagnostics, numerical verification, reproducible provenance, portable scientific export, and optional PyMC, JAX, Lean, and FEniCS/DOLFINx integrations.

v1.12.2 · Model IR · simulation & fitting · causal/Bayesian inference · scientific verification

A production-oriented scientific computing platform built on NumPy, SciPy, pandas, and matplotlib. Its current 5.x line spans 40+ scientific domains and adds guided model fitting with uncertainty, held-out validation, residual diagnostics and reproducibility manifests; compiled C acceleration kernels with safe fallbacks; optional CuPy GPU execution; NumPy Array API support; scikit-learn-compatible estimators; and property, fuzz, oracle, cross-platform wheel, and benchmark-regression testing.

v5.2.4 · 40+ domains · guided scientific fitting · C/GPU acceleration · scientific Python

A zero-runtime-dependency, pure-Python computational science platform whose algorithms are implemented in readable Python rather than hidden behind compiled numerical libraries. It covers numerical methods, linear algebra, statistics, probability, optimization, interpolation, ODE/PDE solvers, signal processing, machine learning, Monte Carlo/MCMC, quantum simulation, symbolic modeling, dimensional/scientific utilities, knowledge tools, and structured falsifiable hypothesis generation, with optional scientific backends available separately.

pure Python · zero runtime dependencies · from-scratch algorithms · numerical science · hypothesis generation

A control layer that decides how an LLM should speak before it speaks.

LLM control · speech gating · Python

Deterministic security reasoning core: hash-chained evidence ledger, attack-path rule engine, fail-closed policy gate. LLMs produce evidence and explain decisions; they never cross the decision boundary.

security reasoning · hash-chained evidence · fail-closed policy

A unified open-source scientific toolkit integrating seven research and AI software projects through the Model Context Protocol (MCP).

MCP · seven-project integration · scientific tooling