FAQ

Questions, answered

Common questions and a glossary so the jargon is never a barrier. For data, the leaderboard, and the starter kit, see Resources.

Questions

Frequently asked

What are the two challenges, and how do the awards map to them?

One award per challenge, matched one to one. Challenge 1 (intra-machine) asks you to reconstruct the flux map on DIII-D, the machine you trained on — the highest composite S_model wins Award #1 ($500). Challenge 2 (cross-machine) asks you to transfer that model zero-shot to MAST — the highest G_ratio wins Award #2 ($500). They share one dataset, one submission format, and one leaderboard (sort by "Ch1: DIII-D S" or "Ch2: G_ratio"). Enter either or both; entering both makes you eligible for both.

Do I need a background in fusion or plasma physics?

No. The starter kit, the dFL visualizer, and four reference baselines are designed so newcomers can reach a first submission without fusion expertise. The data is fully documented, every technical term on this site has a definition you can hover over, and the glossary below spells the rest out.

Can I compete without a GPU?

Yes. Every reference baseline trains to within a few percent of leading numbers in under two hours on a single commodity GPU or CPU. The compute-light badge recognizes accessible solutions. We are also seeking sponsored GPU/cloud credits for teams at resource-constrained institutions — nothing is confirmed yet, and we will post on Announcements if it lands.

What is the compute-light badge?

A self-declared badge for submissions that train end-to-end in under two hours on one GPU or CPU. It is not a separate track, competition, or leaderboard — it applies across both challenges, and there is no compute limit on anyone. It simply marks which entries got there cheaply.

Why is reconstruction without magnetic sensors important?

Reactor-class devices (SPARC, ARC, CFETR) will subject magnetic sensors to neutron fluxes that degrade them severely. Equilibrium reconstruction stays essential, so an inference path from coil currents and electron profiles to ψ(R,Z) becomes a primary pipeline — and a backup for present machines.

What exactly do I submit?

For each test shot and EFIT timestamp: the 65×65 flux map plus just two scalars — q₉₅ and β_N, the only two a flux map cannot contain — one .npz per machine, plus a small manifest.json. Nothing else: the LCFS contour, magnetic axis, shape, volume and l_i are all derived from your flux map by the scorer and scored for consistency with the true flux. Submissions are capped at 5/day and 100 total.

Why are all the plasmas diverted, and why is efit_times not contiguous?

Every timeslice in the release is a diverted plasma — one whose boundary is set by a magnetic X-point. Limited frames, where the boundary is set by physical contact with the wall, were removed on both machines and all splits. The reason is that a limited boundary is determined by wall geometry, which is not contained in ψ(R,Z) at all, so it cannot be inferred from the flux map the challenge asks you to predict. Because those frames were dropped, efit_times is no longer evenly spaced. The full-rate inputs (magnetics, Thomson) are not subset — you still get the complete discharge and resample onto efit_times.

Can I use external data or pre-trained models?

Yes — other public tokamak archives, OMFIT-produced equilibrium tables, and publicly available pre-trained vision or scientific foundation models are all permitted, provided you disclose them in your methods report.

How is cross-machine generalization judged?

Challenge 2 (Award #2) uses G_ratio = S_model(MAST) / S_model(DIII-D) — the fraction of DIII-D performance retained under zero-shot transfer to MAST — among entries reaching a composite S_model ≥ 0.85 on DIII-D (raised from R²ψ > 0.6 in Aug 2026 so the denominator cannot be gamed downward — the scorer enforces the gate and reports G_ratio = 0 below it). Because it is a ratio of the two, it needs predictions for both machines: a DIII-D-only entry scores G_ratio = 0.

When does the data become available, and when are the deadlines?

Training and public test data were released publicly in July 2026 — 9,121 shots, 98 GB. Phase 1 — development runs Jul 27 – Oct 18, 2026 — a continuous public leaderboard, up to 5 submissions a day. Phase 2 — final (blind) runs Oct 19 – Oct 26, 2026 — a blind test set, 3 submissions, leaderboard hidden until it closes. Six sample shots ship in the starter kit for exploration.

Speak the language

Glossary

Every one of these terms is also available as a hover tooltip wherever it appears on the site — you should never have to leave a page to look one up.

Tokamak: A doughnut-shaped machine that confines a fusion plasma with magnetic fields, combining a toroidal (the long way round) and a poloidal (the short way round) field.
Spherical tokamak: A tokamak squashed into more of a cored-apple shape than a doughnut, with the plasma hugging a thin central column. MAST is one; DIII-D is not — which is what makes transferring between them hard.
Shot: A single experimental plasma pulse — one run of the machine, lasting a few seconds. The dataset has one row per shot, and train/test splits are always made by shot, never by timestep.
Poloidal flux ψ(R,Z): The 2D field you are asked to predict. Its contour lines are the magnetic surfaces that hold the plasma in place — so a map of ψ is effectively a picture of the plasma’s shape.
LCFS: Last Closed Flux Surface — the outermost magnetic surface that closes on itself without hitting a wall. In practice: the edge of the plasma.
EFIT: Equilibrium Fitting — the standard code that reconstructs a tokamak’s equilibrium from its sensors. Its output is the ground truth you are scored against; the whole point of the challenge is to reach it without the magnetic sensors EFIT relies on.
Grad–Shafranov equation: The equation governing a tokamak equilibrium. It is the same physics on every machine — which is exactly why a model that has truly learned it should transfer from DIII-D to MAST.
X-point: The point where magnetic surfaces cross in a figure-of-eight. Its presence means the plasma is diverted (exhaust steered to a target); its absence means limited (plasma resting against a wall).
dsep: EFIT-derived context shipped in train and withheld on test — not an input, and not scored. On DIII-D it is EFIT’s a-file DSEP, a separatrix-to-limiter clearance whose sign encodes the configuration (> 0 diverted): it is the criterion that defined this diverted-only corpus. On MAST the same column is a different physical quantity — esm/dr_sep_out, the upper/lower divertor balance — and its sign is not a diverted/limited indicator. Because the two machines’ columns do not mean the same thing, dsep was dropped from the scored metric.
β_N (normalized beta): How much plasma pressure the magnetic field is holding, normalized so machines of different sizes can be compared. One of the two scalars you submit (it depends on the pressure profile, which a flux map cannot contain).
l_i (internal inductance): A measure of how peaked the electric current is across the plasma — high l_i means current concentrated in the core. You do not submit it: the scorer derives l_i from your flux map (part of the Consistency term).
q₉₅ (edge safety factor): How many times a magnetic field line goes the long way round the torus for each time it goes the short way, measured near the plasma edge. It is a stability margin — hence "safety". One of the two scalars you submit (it depends on the toroidal field function, which a flux map cannot contain).
Thomson scattering: A laser diagnostic: fire a laser through the plasma and read the scattered light to get electron temperature (Te) and density (ne). One of the two input families you are given — and, crucially, not a magnetic sensor.
Coil currents: The currents in the machine’s electromagnets — the "control panel" of the tokamak. The other input family. DIII-D has 18 F-coils plus ECOILA and bcoil; MAST has 10 P-coils plus solenoid, TF and EFPS.
Ip (plasma current): The total electric current flowing through the plasma itself — of order 10⁶ A, roughly 100× the coil currents. Given as an input.
S_model: The composite score that decides Challenge 1: S = 0.55·R²ψ + 0.15·R²_q95,βN + 0.10·(1 − D_LCFS) + 0.20·Consistency. It blends flux-map accuracy, the two predicted scalars, boundary alignment, and the consistency of the eight ψ-derived scalars into one number in [0, 1]. Higher is better.
G_ratio: The score that decides Challenge 2: G = S_model(MAST) / S_model(DIII-D) — the fraction of your DIII-D performance you keep when moving zero-shot to MAST. 1.0 would mean perfect transfer. It needs predictions for both machines, and only entries with S_model(DIII-D) ≥ 0.85 are eligible — below that gate (or with DIII-D missing) the scorer reports 0.
D_LCFS: How far your predicted plasma boundary sits from the true one — a symmetric Hausdorff distance, normalized by the true boundary’s major radius. Smaller is better. You do not submit a contour; the scorer extracts it from your flux map.
Zero-shot: Applying a model to a new machine with no training data from it at all — no fine-tuning, no examples. MAST has zero training shots on purpose; that is the whole of Challenge 2.
R²: Coefficient of determination, 1 − SS_res/SS_tot. 1.0 is perfect; 0 is no better than predicting the mean.
SSIM: Structural Similarity Index — an image-quality metric that rewards getting the structure right rather than each pixel exactly. Used to describe baseline results, not to score submissions.
Hausdorff distance: The worst-case gap between two shapes: how far you must travel from the least-well-matched point on one contour to reach the other. Used to compare predicted and true plasma boundaries.
PCA: Principal Component Analysis — the recommended first move: compress the 4,225-pixel flux map down to ~20–50 numbers (the first component alone carries ~92% of the variance), predict those, then reconstruct.
Compute-light badge: A self-declared badge for models that train end-to-end in under two hours on a single GPU or CPU. It is a recognition that applies across both challenges — not a separate track, and not a separate leaderboard.
Codabench: The open-source platform hosting the leaderboard and scoring your submissions.
dFL: Data Fusion Labeler — the free desktop app that renders any shot in the dataset (flux contours, Thomson profiles, coil traces) so you can eyeball your inputs before you train.