Update dataset card

README.md

@@ -35,6 +35,11 @@ variables on a 32 x 64 x 10 latitude-longitude-pressure grid. It includes three
 nested benchmark subsets, two evaluation protocols, and eight released baseline
 methods.
 
+Inputs are 8 continuous planet parameters plus the source GCM label. Outputs
+are time-averaged climate fields on a 32 x 64 latitude-longitude grid:
+three-dimensional variables are stored as pressure-level channels, and
+two-dimensional variables are stored as single fields.
+
 ![ThousandWorlds dataset schematic](OVERVIEW.png)
 
 ## Quickstart
@@ -65,25 +70,30 @@ and reproducibility instructions.
 
 ## Files
 
-The release is distributed as archives so that the on-disk layout matches the
-ThousandWorlds package directly.
+The release includes:
 
-| File | Contents |
-| --- | --- |
-| `archives/dataset.tar.gz` | Metadata CSVs, benchmark subsets, gridded climate fields, spectral coefficients, and normalization assets. Extracts to `dataset/`. |
-| `archives/results-baselines-<subset>-deterministic.tar.gz` | Deterministic baseline predictions, configs, metrics, and tables for one subset. Extracts to `results/`. |
-| `archives/results-baselines-<subset>-gplfr.tar.gz` | GPLFR probabilistic baseline predictions and metrics for one subset. Extracts to `results/`. |
-| `archives/results-baselines-<subset>-ppca_icm.tar.gz` | PPCA-ICM probabilistic baseline predictions and metrics for one subset. Extracts to `results/`. |
-| `archives/*.sha256` | SHA256 checksums for the release archives. |
-| `croissant.json` | Machine-readable Croissant metadata for the archive layout. |
+- `archives/dataset.tar.gz`: the ThousandWorlds dataset.
+- `archives/results-baselines-*.tar.gz`: baseline predictions for the 3
+  subsets.
+- `croissant.json`: Croissant metadata.
+- `archives/*.sha256`: checksum sidecars.
+
+## Dataset Contents
+
+The dataset contains gridded fields (NumPy), input metadata (CSV), predefined
+train/test splits, normalization statistics, and spherical harmonic
+coefficients plus inverse-SHT weights for spectral methods.
+
+## Subsets
 
-Canonical subsets are:
+The dataset is organized into three subsets of increasing complexity and
+realism:
 
 | Subset | Simulations | Fields | Description |
 | --- | ---: | ---: | --- |
-| `single-complete` | 256 | 48 | UM-only complete-observation subset |
-| `multi-complete` | 1659 | 48 | 5 GCMs, complete-observation subset |
-| `multi-partial` | 1760 | 53 | 5 GCMs, subset with structured whole-field missingness |
+| `single-complete` | 256 | 48 | Smaller subset; simulations from a single GCM, complete observations only. |
+| `multi-complete` | 1659 | 48 | All 5 GCMs, still with no missing fields. |
+| `multi-partial` | 1760 | 53 | Full dataset; all 5 GCMs, with missing fields represented as NaNs. |
 
 The subset split files contain:
 
@@ -151,11 +161,12 @@ and LFRic.
 
 ## Evaluation
 
-The package includes loaders and metrics for the two benchmark protocols:
+The package includes loaders and metrics for two benchmark protocols:
 
-- **Standard**: evaluate on each subset's held-out test simulations.
+- **Standard**: the main test protocol, ideal for ML model comparison.
 - **Shared-planets**: evaluate on planets shared across target and auxiliary
-  GCMs; available for the two multi-GCM subsets.
+  GCMs; used to assess performance relative to inter-GCM error, i.e. how close
+  a model gets to the epistemic uncertainty floor of the problem.
 
 Released baselines include train mean, kNN, PCA ridge, PCA-MLP, Coord-MLP,
 Coord-DeepONet, PPCA-ICM, and GPLFR. Baseline artifacts include predictions,