v0.23.0 — 17 July 2026 Current
Configurable Coupling & Modular Runtime Hardening
Exposed validated gyrokinetic-to-UPDE coupling gains, added graceful phase-stream shutdown, and split evidence validation, transport geometry, runtime sanitisation, multi-ion evolution, auxiliary heating, and anomalous transport into explicit modules while preserving the existing CLI and numerical contracts. Current generated inventory: 196 source modules, 64 Rust modules, 155 validation scripts, and 546 Python test files.
v0.22.0 — 26 June 2026
Fuzzing, Benchmark-Regression Gate & Runtime-Bound Safety Certificate
Added a libFuzzer harness for the Rust parser, numeric-adapter, and FFI surfaces (reactor-configuration JSON, VMEC-like and BOUT++ text parsers, the capacitor-bank discharge ledger, and the Kuramoto phase kernel), with a fail-closed campaign orchestrator, a tracked seed corpus, and a nightly fuzzing workflow separate from per-commit CI. Added a polyglot benchmark regression gate that records per-language p50/p95/p99 latency, throughput, and run provenance and fails closed on tampered reports, missing metrics, or cross-CPU comparisons. Added a runtime-bound formal safety certificate that binds a holding bounded CTL/LTL certificate to a structured controller runtime identity (controller configuration, Petri-net topology digest, SNN parameters, solver mode, and a declared timing envelope) — issuance, proof replay, and facility-facing admission all fail closed unless the binding re-proves on the declared stack. Fixed an overflow in the Rust matrix-exponential scaling-and-squaring exponent that could stall the capacitor-bank discharge (it now fails closed on a non-finite norm). Capability inventory: 146 Python modules, 64 Rust source files, 8,215 tests.
v0.21.0 — June 2026
Reference-Validation Release
Validated the Kuramoto phase-synchronisation runtime against published results, the Grad-Shafranov equilibrium solver against the Solov'ev exact solution, and the structured-singular-value (μ) computation against exact μ references. Capability inventory: 145 Python modules, 64 Rust source files, 6,257 tests.
v0.20.0 — June 2026
Release Preparation & Documentation
Reworked the README, documentation landing page, onboarding guide, use-case page, and tutorials; regenerated the capability manifest and README snapshot from the current tree. Patch series v0.20.1–v0.20.7 followed with incremental physics and packaging fixes.
v0.19.0 — March 2026
33 Physics Corrections & 100% Coverage
Major audit release. 33 equation corrections across GS solver, MHD stability criteria, neoclassical transport, and edge models. All implementations verified against original publications. 3,300+ tests, 100% line coverage. Sauter bootstrap current coefficients corrected. Mercier criterion sign convention fixed. Ballooning eigenvalue boundary conditions revised. L-H Martin 2008 exponents verified.
v0.18.0
Nonlinear δf Gyrokinetics
5D Vlasov equation solver for nonlinear δf gyrokinetics. Orszag 2/3 dealiasing rule. Dimits shift verified against GS2/GENE reference. Zonal flow generation and saturation. 53 new tests covering convergence, conservation laws, and benchmark cases.
v0.17.0
GK Three-Path System & AGPL
Gyrokinetic transport reorganised into three-path system: critical-gradient (fast), QLKNN surrogate (medium), native eigenvalue (accurate). Licence changed from MIT to AGPL-3.0. 163 new tests. SAT0/SAT1/SAT2 saturation models for TGLF-equivalent tier.
v0.16.0
Phase 3 & 4: Frontier Physics + Absolute Control
H∞ and μ-synthesis controllers. Sliding-mode vertical stability. Fault-tolerant control with innovation monitoring. EPED pedestal model. SOL two-point model. MARFE radiation collapse. NTM Modified Rutherford equation. Tearing and locked mode physics. ELM peeling-ballooning boundary.
v0.14.0
PPO Cloud Training
PPO reinforcement learning agent trained for 500K steps on Vertex AI T4. Gymnasium-compatible tokamak environment. Reward 143.7 vs MPC 58.1 vs PID -912.3. 0% disruption rate over 100 evaluation episodes. Checkpoint export and inference-only mode.
v0.13.0
JAX-Differentiable Grad-Shafranov
Full GS equilibrium solve under JAX autodiff. JIT compilation. GPU support. Gradient through complete equilibrium reconstruction. Neural PCA+MLP surrogate (1000× speedup). Enables gradient-based plasma shape optimisation.
v0.12.0
QLKNN-10D + PPO Agent
QLKNN-10D neural network surrogate for gyrokinetic transport (ion + electron channels). PPO agent initial implementation with basic tokamak environment. MPC controller with 20-step horizon. Digital twin framework foundation.