diff --git a/.claude/skills/add-benchmark/SKILL.md b/.claude/skills/add-benchmark/SKILL.md
index 385666e384..d2b0cb65e4 100644
--- a/.claude/skills/add-benchmark/SKILL.md
+++ b/.claude/skills/add-benchmark/SKILL.md
@@ -72,7 +72,7 @@ Convert your source dataset to Gym JSONL format. Each line must have `responses_
 
 **Data conversion**: Write conversion scripts in the **source repo** (e.g. your dataset repository), not in NeMo-Gym. Prompt files also belong in the source repo. Exception: when there is no external source repo. See `references/patterns.md` § "Data Conversion Script Pattern".
 
-**`example.jsonl`**: Generate 5 entries for smoke testing. This file is committed directly to git in `data/example.jsonl`.
+**`example.jsonl`**: Generate 5 entries for smoke testing, committed directly to git in `data/example.jsonl`. Each row must be in materialized Responses format (contain `responses_create_params`) — example datasets are not templated via `prompt_config`, so raw rows won't validate.
 
 **`train`/`validation` datasets**: Upload to the GitLab dataset registry — these must NOT be committed to git.
 
diff --git a/.claude/skills/add-benchmark/references/patterns.md b/.claude/skills/add-benchmark/references/patterns.md
index 68d3668b9d..5a56bf8156 100644
--- a/.claude/skills/add-benchmark/references/patterns.md
+++ b/.claude/skills/add-benchmark/references/patterns.md
@@ -247,8 +247,7 @@ my_benchmark_eval_agent:
 ### Key rules
 
 - The `verified: false` flag is auto-added by pre-commit hook. Set to `true` after baselining.
-- `license` is required for `train` and `validation` datasets.
-- Valid license values: `Apache 2.0`, `MIT`, `CC-BY-4.0`, etc.
+- `license` is required for `train` and `validation` datasets. It must be one of the values accepted by the dataset config schema (validated enum in `config_types.py`).
 - `domain` should be one of: `coding`, `math`, `other`, or check `config_types.py` for current enum.
 - Dataset `type` must be one of: `train`, `validation`, `example`.
 - `gitlab_identifier` is required for `train`/`validation` datasets. `jsonl_fpath` is the local download path. Both fields coexist.
diff --git a/fern/versions/latest/pages/data/prepare-validate.mdx b/fern/versions/latest/pages/data/prepare-validate.mdx
index 3c024e6e95..528f1b4299 100644
--- a/fern/versions/latest/pages/data/prepare-validate.mdx
+++ b/fern/versions/latest/pages/data/prepare-validate.mdx
@@ -239,6 +239,14 @@ ng_prepare_data "+config_paths=[resources_servers/example_multi_step/configs/exa
     +mode=example_validation
 ```
 
+<Info>
+**Resources server example-data contract (PR submission).** A `resources_servers/<name>/` passes data validation when its `data/` directory contains `example.jsonl`, the `example_metrics.json` produced by `example_validation`, and `example_rollouts.jsonl`. Commit all three.
+</Info>
+
+<Warning>
+`example.jsonl` must already be in materialized Responses format (each row containing `responses_create_params`). Example-type datasets are not templated via `prompt_config`, so raw rows won't validate.
+</Warning>
+
 ### Training Preparation
 
 ```bash
@@ -378,6 +386,10 @@ datasets:
 | `train` | Training data | RL training |
 | `validation` | Evaluation during training | RL training |
 
+### License values
+
+`license` is required for `train` and `validation` datasets. It is validated against an allowed set defined in code — see the `license` enum in [`nemo_gym/config_types.py`](https://github.com/NVIDIA-NeMo/Gym/blob/main/nemo_gym/config_types.py).
+
 ---
 
 ## Next Steps