Release History

Release History#

Level 2 is regenerated when we add data or change how it is processed, replacing files at the same paths, so the same shot downloaded at different times can differ. Changes are listed newest first, with what older files contain. For details, see Changes From Source.

Which version do you have?#

Each level 2 group records when it was produced and from which ingestion commit:

import xarray as xr

ds = xr.open_zarr("https://s3.echo.stfc.ac.uk/mast/level2/shots/30421.zarr", group="summary")
print(ds.attrs["ingested_at"], ds.attrs["commit_url"])

Release

Ingested

Includes

Corrections release

TBC

Everything on this page

September 2026

23–29 September 2026, commit ed6ea4e

Changes up to 24 August 2026

Corrections release#

Changes from source#

  • CC04 withheld from shot 25366. Older files: reversed (25470–25793), then flat (25794 onwards). See magnetics.

  • Saddle coils M02, M08, M012 sign-corrected. Older files: field reversed from 24782, voltage reversed up to 24781. See magnetics.

  • PF scaling completed (SOL ×0.5, P6L/P6U ×4). September 2026 files: the solenoid is 2× too high and P6 is 4× too low. See pf_active.

Fixes to our own processing#

Samples dropped or misplaced on the time grid#

Affected: grid-resampled datasets, most visibly equilibrium, magnetics, summary, bolometer and the 50 µs datasets.

Level 2 matched sample times to grid points by exact comparison, which dropped or misplaced samples, especially near t = 0. On shot 30421, 18 of 107 wmhd slices held the previous slice’s value and two EFIT solutions were dropped (131 of 141 present after the fix, 129 before). On shot 26399, only 2,293 of 4,200 pf_active grid points held their own sample; now all do.

Matching now uses the sampling step (#152). Unaffected datasets are unchanged.

Thomson scattering time axis doubled#

Affected: 124 of the 1,177 M9 shots with Thomson data: 30073–30210 (almost all), 29525, 29574–29575, 30040, 30321, 30417–30419, 30421, 30441 and 30443.

The core (t_e_core, n_e_core) and profile (t_e, n_e, p_e) signals have slightly different time stamps for the same laser pulses. Merging kept both, giving 283 time steps for 145 pulses on shot 30421, half of them empty for any variable.

Axes sampled at the same rate are now merged, one step per pulse (#152). Empty values on shot 30421 fall from 57.6% to 16.4%; the rest are channels outside the 0.30–1.49 m grid. Genuine burst-mode gaps are unaffected: they show in a single signal, while this fault only appeared after merging.

Saddle coil voltage channels empty#

Affected: the last three channels of b_field_tor_probe_saddle_voltage, shots 11708–27619.

Level 2 requested M10–M12, but the archive names them M010–M012, so they were all-NaN. Now read correctly (#157).

Earlier changes#

Dates are merge dates. Correction changes values taken from the archive; Processing changes how level 2 is produced.

Merged

Type

Change

Older files

2026-08-24

Processing

Profiles split by shot range take attributes from the right range (#144)

units and description may come from another part of the campaign

2026-08-24

Processing

EFIT constraint axes named via n_flux_loop_order (#143)

Unnamed numbered axes

2026-08-18

Processing

Thomson radial grid ends at 1.49 m (17f6465)

121 radial points, not 120

2026-08-18

Processing

Thomson uses the laser’s pulse times (#142)

Uniform 5 ms grid that smoothed fast events, such as a tearing mode on burst-mode shot 24623

2026-07-15

Correction

PF feed currents multiplied by turns (#129); also dropped the solenoid’s ×0.5

Per-turn currents; solenoid halved

2026-07-14

Processing

Saddle field channels M010–M012 read correctly (#127)

All-NaN

2026-07-10

Processing

shot_id is a coordinate, not a dimension (#124)

Extra length-one shot_id dimension

2026-06-01

Processing

Mistyped Thomson time step fixed to 5 ms (#120); superseded by #142

Step of 5×10⁻³⁵ s

2026-05-22

Processing

shot_id added, so shots can be concatenated (#98)

No shot_id

2026-05-15

Processing

All time grids start at −0.1 s (#111)

Start follows plasma onset, so axes do not align across shots

2026-05-15

Processing

ingested_at and commit_url attributes added (#113)

No provenance

2026-05-01

Processing

Channel coordinates named consistently after UDA signals (#107)

Inconsistent channel names

2026-04-27

Correction

Solenoid halved and moved into pf_active.coil_current as SOL (#110)

Separate pf_active.solenoid_current, 2× too high

2026-04-27

Processing

A signal of unexpected length no longer drops the rest of its source (#102)

Data may be missing because of one such signal

2026-03-20

Processing

Corrupted shots 26958, 26963, 26964, 26967 re-ingested (issue #105)

Unreadable

2025-12-18

Processing

Dataset attributes no longer copied from a variable (#97)

For example, summary has units: A from ip

2025-09-08

Processing

IMAS descriptions added where they exist (#88)

No descriptions

2025-06-24

Processing

X-point channel names and size fixed (#83)

Different names and size

2025-05-19

Processing

Valve and target voltages published, with valve group attributes (#72); see gas_injection

No valve data

2025-05-01

Processing

pf_active current and voltage get separate channel dimensions (#71)

Both use channel

2025-04-08

Processing

Time is the last dimension in charge_exchange, thomson_scattering, equilibrium, camera_visible (#67)

Inconsistent dimension order

2025-04-02

Correction

Saddle voltage offsets subtracted (#55)

Offsets included