Issues

• data: observation products for scoring in observation space, including the deep ocean

  enhancement
  m2lines/Samudra#996 · Filter by author alxmrs opened on Nov 14, 2026

• data: evaluation products for an LLC4320 scorecard track

  enhancement
  m2lines/Samudra#995 · Filter by author alxmrs opened on Nov 14, 2026

• data: publish scorecard reference products (climatology, anomaly persistence, layer weights) with each OM4 release

  enhancement
  m2lines/Samudra#994 · Filter by author alxmrs opened on Nov 14, 2026

• Scorecard beyond OM4: track definitions, LLC4320, observations, and the deep ocean

  enhancement
  m2lines/Samudra#993 · Filter by author alxmrs opened on Nov 14, 2026

• Publish the scorecard on the docs site, built in CI from a git registry and OSN

  enhancement
  m2lines/Samudra#992 · Filter by author alxmrs opened on Nov 14, 2026
  1

• jaxtyping shape annotations in samudra are never checked at runtime

  m2lines/Samudra#990 · Filter by author alxmrs opened on Nov 13, 2026

• Scorecard eval: skill against references and the champion, with regression gates

  m2lines/Samudra#986 · Filter by author alxmrs opened on Nov 13, 2026
  3

• Experiment: compression-baselines ladder at 1/4°, on the #956 classifier task

  experiment
  m2lines/Samudra#985 · Filter by author alxmrs opened on Nov 13, 2026

• docs/utils/loss.md states the dynamic loss weighting direction backwards

  bugdocumentation
  m2lines/Samudra#982 · Filter by author alxmrs opened on Nov 13, 2026

• Losses that score whole fields, not cells: measured case and plan

  enhancement
  m2lines/Samudra#981 · Filter by author alxmrs opened on Nov 13, 2026
  1

• Property-based and metamorphic tests for ocean state invariants

  enhancement
  m2lines/Samudra#977 · Filter by author alxmrs opened on Nov 12, 2026

• Publish reference statistics (Parquet) with each data release and validate new data against them

  enhancement
  m2lines/Samudra#976 · Filter by author alxmrs opened on Nov 12, 2026

• OHC uses two heat capacities: 3992 (constants) vs 3850 (metrics), a 3.7% disagreement

  bug
  m2lines/Samudra#975 · Filter by author alxmrs opened on Nov 12, 2026

• Confirm near-fresh salinity cells (min 0.03 psu) in v2026-09 OM4 are physical estuaries

  question
  m2lines/Samudra#974 · Filter by author alxmrs opened on Nov 12, 2026

• v2026-09 OM4 ocean volume is ~10% high: bottom cells count full dz, no bathymetry in stores

  bug
  m2lines/Samudra#973 · Filter by author alxmrs opened on Nov 12, 2026

• areacello in v2026-09 stores: mislabeled units (km2 → m2) and polar Earth radius (0.47% low)

  bug
  m2lines/Samudra#972 · Filter by author alxmrs opened on Nov 12, 2026

• Latent rollout experiments via JEPAs

  m2lines/Samudra#968 · Filter by author alxmrs opened on Nov 9, 2026
  

• Fix loss weighting to use wet cells as denominator, not total cells

  m2lines/Samudra#964 · Filter by author jder opened on Nov 6, 2026
  5

  

• Preserve depth information in OM4 export

  m2lines/Samudra#961 · Filter by author jder opened on Nov 5, 2026
  2

• Experiment: exclude timestamp-gap-crossing windows (PR #946)

  m2lines/Samudra#953 · Filter by author oa-jder-bot opened on Oct 24, 2026

• Experiment: detach gradients between autoregressive steps

  m2lines/Samudra#952 · Filter by author oa-jder-bot opened on Oct 24, 2026

• Train Train

  m2lines/Samudra#950 · Filter by author oa-jder-bot opened on Oct 24, 2026
  1

• Reprocess om4 averages data from source netcdf files to include wfo

  m2lines/Samudra#937 · Filter by author jder opened on Oct 19, 2026
  3

  

• Look at latitude padding artifact

  m2lines/Samudra#923 · Filter by author jder opened on Oct 13, 2026
  2

• Training on my own data

  m2lines/Samudra#922 · Filter by author RowenXu opened on Oct 12, 2026
  2