kronos-ml / CODES.md
Kronos Fusion Energy
Refresh to v0.3.0 (39 codes); clean package + docs (drop stale cache dump and deposit tooling)
2573a5a
|
Raw History Blame Contribute Delete
14.1 kB

KODEX codes — reference

One section per code: brand, technical function, rollout phase, status, the real code it stands in for, and provenance. Full benchmarks in BENCHMARKS.md.

Phase 1 — data-backed core (ships first)

KAIROS — CONTROL [BUILT]

real-time closed-loop MPC control; supplies the model-free L4 clamp that KGATE enforces. provenance TWIN · retired_by validated real-time plant controller (device-in-the-loop) · gates AC-24, AC-25.

KFLOW — STATE [BUILT]

twin state-estimation / dropped-sensor imputation on the 76-node diagnostic sensor graph. provenance SIM · retired_by trained message-passing GNN (gnn_seed0..4.pt) · gates AC-18, AC-19.

KGATE — SAFE [BUILT]

trust-boundary / abstention gate — a model-free envelope clamp that decides in_domain and fail-closes on bad input. provenance TWIN · retired_by formal CBF verification + hardware failsafe.

KHALO — UQ [BUILT]

shared calibrated-uncertainty layer (GP posterior + deep-ensemble) + the canonical reliability report. provenance SIM · retired_by full ensemble UQ / conformal prediction · gates KX-L1-A14, KX-L1-A5.

KMAT — MATERIALS [BUILT]

ML+DFT alloy / HEA screen (CHGNet) for structural & functional materials; screen BUILT, dpa/lifetime roadmap. provenance SIM · retired_by DFT (VASP / Quantum ESPRESSO) · gates KX-L1-A4, BR-L2-A8.

KOIL — MAGNET [BUILT]

magnet-twin stress-field surrogate + quench-precursor early-warning; REBCO per-turn Wilson-form hoop stress. provenance SIM · retired_by full FE electromechanical solve · gates BR-MT1, BR-MT3.

KORE — EQUIL [BUILT]

fast MHD-equilibrium accelerator (Fourier Neural Operator / DeepONet); Grad-Shafranov, not turbulence. provenance ANALYTIC · retired_by classical Grad-Shafranov solver · gates AC-16, AC-L1.

KQUBIT — QML [BUILT]

REAL VQE (PennyLane) — variational quantum eigensolver for a molecular (H₂) Hamiltonian; runs on a simulator now and on real quantum hardware (KODEX_QC_BACKEND=ibm + token), same code path. Recovers the exact ground state to 0.0002 mHa; a bundled depolarizing-noise sweep shows p=0.5% → 5.56 mHa (why there's no advantage yet). Honest: no quantum advantage this decade (~8–10 yr out); the tooling is real today. provenance SIM · retired_by fault-tolerant quantum hardware (not available this decade) · gates AC-44, KX-L3-A4.

KWARD — DISRUPT [BUILT]

disruption / stability-boundary early-warning (soft-vote ensemble); advisory, downstream of the KGATE clamp. provenance TWIN · retired_by real-device disruption-labelled training (C-Mod/DIII-D/EAST) · gates AC-20, AC-21.

KYRO — TRANSPORT [BUILT]

CGYRO turbulence-transport surrogate (ion heat flux Q_i vs a/L_T x shear); representative fidelity mu=400, real-mass gold pending. provenance CGYRO-urep · retired_by CGYRO (nonlinear gyrokinetic) · gates BR-L2-A12, BR-L2-A1e, BR-L2-A1c.

Phase 2 — fleet extensions

KBREED — breeder [BUILT]

TBR / neutronics / isotope-yield surrogate for the Hyperion breeder. provenance SIM · retired_by OpenMC / high-fidelity neutronics · gates BR-neut.

KBURN — burner [BUILT]

net-power balance / dispatch / plant surrogate (Aegis/MetroVolt). provenance SIM · retired_by full plant systems model.

KEYE — DIAG [BUILT]

diagnostics / virtual-sensor / signal-validation observer — builds on KFLOW's graph imputation + quantum-sensing; flags faulty channels. provenance SIM · retired_by physical diagnostic suite + validated sensor fusion · gates AC-36.

KFLUX — neutronics [BUILT]

neutronics surrogate — shielding / activation / flux maps. provenance SIM · retired_by OpenMC.

KFUSE — MULTIFID [BUILT]

multi-fidelity surrogate fusing reduced-twin + mu=400 CGYRO + (pending) real-mass gold, with fidelity-aware uncertainty. provenance CGYRO-urep · retired_by CGYRO real-mass converged (mu=3672) · gates BR-L2-A1e, BR-L2-A1c-MS.

KHEAT — heating&CD [BUILT]

heating & current-drive actuator-response surrogate — RandomForest over a 3888-point current-drive design scan; predicts driven current I_cd from RF/NBI drive parameters + plasma state (R²=0.949). Feeds KAIROS heating control. Reduced CD model — RF/NBI ray-tracing is the fidelity upgrade. provenance SIM · retired_by full RF/NBI ray-tracing (GENRAY/TORAY/NUBEAM) · gates H10.

KISO — isotopes [BUILT]

isotope-production surrogate incl. medical isotopes. provenance SIM · retired_by activation transport + separation model.

KPATH — operational [BUILT]

pulse-trajectory optimizer (ramp-up -> flat-top -> ramp-down). provenance SIM · retired_by operational scenario optimizer.

KQROSS — QRE [BUILT]

REAL FT resource estimator — computes the classical↔quantum crossover, surface-code physical-qubit overhead, and logical-qubit roadmap year for a fusion electronic-structure kernel (order-of-magnitude, literature-scaled). Crossover N=50 → **6e4 physical qubits**, roadmap ~2032; today's hardware ~1e2–1e3 physical, no logical. Validated negative: no FT advantage this decade. provenance SIM · retired_by fault-tolerant quantum hardware (not available this decade) · gates AC-43, BR-SX-08.

KSENSE — QSENSE [BUILT]

REAL quantum-metrology (PennyLane mixed-state GHZ) — quantum Fisher information: ideal GHZ gives Heisenberg scaling (gain ~N), but realistic NV/SQUID/SERF dephasing collapses it to the standard quantum limit (over N=2→8, ideal grows 4× vs dephased 2.05× = SQL rate). Honest null: no quantum-sensing advantage for fusion disruptions this decade. provenance SIM · retired_by deployed physical diagnostic hardware · gates HX-29.

KTENSOR — TN [BUILT]

tensor-network / low-rank (MPS-style SVD) ROM of the real CGYRO A1e flux database. Honest measured result: the flux is only modestly compressible — effective rank ~2.93 of 4; bond-dim-2 keeps ~91% of the variance but ~30% rel-L2, and the turbulent↔quiet transition is a sharp high-rank feature. A genuine ROM characterization that corrects the naive "flux is trivially low-rank" expectation. Classical SVD/MPS — no quantum-advantage claim. provenance SIM · retired_by exact many-body / quantum simulation · gates KX-L3.

KQERN — QKERNEL [BUILT] ✦v0.2

REAL quantum kernel (PennyLane) — embeds real MAST disruption features into a quantum feature map and classifies on the fidelity kernel; runs on a simulator now, real QC hardware pluggable. MAST-disruption AUC 0.919 vs classical RBF 0.929 (ties — a validated null); a real, runnable quantum-ML pipeline on real fusion data. provenance SIM · retired_by fault-tolerant quantum hardware (not available this decade) · gates KX-L3.

KMIT — QMITIGATION [BUILT] ✦v0.3

REAL quantum error mitigation — zero-noise extrapolation (ZNE, global unitary folding) + inverse-confusion readout-error mitigation applied to the real KQUBIT H₂ VQE circuit (PennyLane default.mixed; sim now, real hardware pluggable). ZNE removes 89% of the noise-induced energy error (44.4 → 5.0 mHa); readout mitigation cuts the measurement TV distance to ~0. Mitigation is the genuine NISQ tool TODAY — it recovers signal from noise, not a speedup. Honest: quantum's meaningful role in fusion is post-2036. provenance SIM · retired_by fault-tolerant quantum hardware (not available this decade) · gates KX-L3.

Phase 3 — cutting-edge + capstones

KDRIVE — RL-control [BUILT]

reinforcement-learning plasma controller (learned policy). provenance n/a · retired_by validated RL controller.

KECON — techno-economics [BUILT]

GENERIC open techno-economics (LCOE) on the USER's inputs — FINANCIAL FIREWALL: no Kronos numbers, ever. *provenance n/a · retired_by detailed engineering cost model (never public).*

KEDGE — edge-transport [BUILT] ✦v0.2

divertor / edge heat-flux → target-thermal surrogate — maps divertor heat flux to target surface temperature + material limits (W / CuCrZr) from the H9 exhaust scan. q→T_surf fit R²=1.0 (77 pts); CuCrZr material limit q~13.5 MW/m². Reduced 0-D thermal model — SOLPS-ITER/EIRENE edge campaign is the fidelity upgrade. provenance SIM · retired_by SOLPS-ITER / EIRENE edge campaign · gates H9.

KFORGE — inverse-design [BUILT]

inverse-design optimizer that chains the whole fleet to search designs. *provenance n/a · retired_by the fleet itself (capstone).*

KFUEL — fuel-cycle [BUILT]

tritium fuel-cycle & inventory surrogate. provenance n/a · retired_by fuel-cycle systems code.

KGEN — generative [BUILT]

generative / diffusion surrogate for states, equilibria & fields. provenance n/a · retired_by validated generative model.

KLAW — eqn-discovery [BUILT]

equation discovery (sparse / symbolic regression, SINDy-class). provenance n/a · retired_by verified reduced closures.

KPILOT — agentic-AI [BUILT]

agentic AI copilot orchestrating the fleet in natural language. provenance n/a · retired_by the fleet + an agent runtime.

KRAD — radiation-control [BUILT] ✦v0.2

impurity-seeding radiation evaluator — maps a seeded impurity + radiated-power fraction to core Zeff penalty and P_rad from the H9 seeding scan (N / Ne / Ar; Ar gives the least core Zeff dilution at f_rad 0.58). Small scan; full impurity-transport (SOLPS + impurity) is the fidelity upgrade. provenance SIM · retired_by impurity transport (SOLPS + impurity) + radiation control · gates H9.

KQOPT — QOPT [BUILT] ✦v0.2

REAL QAOA optimizer (PennyLane) — solves a combinatorial design QUBO on a real quantum circuit; plug in your own QUBO. Runs on a simulator now, real QC hardware pluggable. Reaches 100% of the brute-force optimum on a 6-qubit design QUBO (p=2). No speedup vs classical at this size; advantage ~8–10 yr out. provenance SIM · retired_by fault-tolerant quantum hardware (not available this decade) · gates KX-L3.

KDYN — QDYN [BUILT] ✦v0.2

REAL quantum dynamics (PennyLane) — Trotterized time-evolution of a transverse-field Ising "kinetic" Hamiltonian; runs on a simulator now, real QC hardware pluggable. 1st-order Trotter error falls 1.35→0.034 over 1→32 steps (~1/n) — the honest cost curve for simulating plasma-like dynamics on a quantum computer. provenance SIM · retired_by fault-tolerant quantum hardware (not available this decade) · gates KX-L3.

KSEEK — ACTIVE [BUILT] ✦v0.2

active-learning acquisition — proposes the next most-informative CGYRO run from KYRO's GP posterior (max-variance / uncertainty sampling), so expensive GPU-hours go where the surrogate is least sure. Honest: leave-one-out std↔error correlation is weak (−0.28) on the 16-point map — directional guidance that strengthens with data. provenance SIM · retired_by the full CGYRO parameter scan · gates BR-L2-A1e.

KBENCH — BENCHMARK [BUILT] ✦v0.2

open, citable fusion-ML benchmark suite — 3 tasks (CGYRO turbulence, MAST disruption, flux ROM) with real on-disk data, fixed reproducible splits, metrics, and KODEX baselines to beat. A community leaderboard starting point, not a private result; honest that it's small by mainstream-ML standards (pilot fusion-ML benchmarks). provenance n/a · retired_by community-standard fusion-ML benchmarks.

KMIND — FOUNDATION [BUILT] ✦v0.3

whole-device foundation surrogate — ONE shared-trunk multi-task neural net (per-domain input adapters → a shared plasma-state trunk → per-domain heads, ~67k params) trained JOINTLY across transport (KYRO), equilibrium (KORE), state (KFLOW) and disruption (KWARD) on the data those specialists use. Benchmarked head-by-head against a MATCHED single-task baseline on the same split: on the well-populated domains the shared representation shows no negative transfer (equilibrium slight positive, state/disruption neutral); transport is data-starved (16 pts) and reported, not hidden. MC-dropout uncertainty; per-domain trust gate (physical box for transport/equil, Mahalanobis for state/disrupt). The honest question a foundation model must answer — does one shared representation help or hurt each task? — answered with matched baselines. provenance SIM · retired_by a device-scale multi-task foundation model on real multi-diagnostic data · gates BR-L2-A1e, AC-16, AC-18, AC-20.

KLINQ — QLINSOLVE [BUILT] ✦v0.3

REAL quantum linear solver — a variational quantum linear solver (VQLS) prepares the solution of a 4×4 reduced-MHD (discrete-Laplacian, SPD) linear system with a real quantum circuit (PennyLane; sim now, hardware pluggable), recovering the classical solution at fidelity 1.0. Bundled with an honest fault-tolerant HHL resource estimate: a fusion-scale (N1e6) sparse solve needs 4e4 physical qubits — no advantage at real scale. Honest: quantum's meaningful role in fusion is post-2036. provenance SIM · retired_by fault-tolerant quantum hardware (not available this decade) · gates KX-L3.

KSCOUT — MF-CAMPAIGN [BUILT] ✦v0.3

multi-fidelity Bayesian-optimization campaign manager — extends KSEEK from a single next-point to a budget-constrained campaign that chooses which operating point AND which fidelity (reduced twin / μ=400 CGYRO / real-mass gold) to run next. Triage-then-escalate: cheap twin on every uncertain hotspot, escalate the most-uncertain to μ=400, reserve real-mass gold. Under a tight GPU-hour budget it reaches ~2.3× the mean truth-confidence of a naive all-μ=400 plan (which exhausts the budget and leaves most hotspots unexamined). Honest: fidelity correlations are assumptions (gold not yet run); GPU-hours only, no cost figures. provenance SIM · retired_by the full multi-fidelity CGYRO campaign (twin + μ=400 + real-mass gold) · gates BR-L2-A1e.