diff --git a/docs/applications/kegg_curation_implementation_status.md b/docs/applications/kegg_curation_implementation_status.md
index 647c0a2..c434c78 100644
--- a/docs/applications/kegg_curation_implementation_status.md
+++ b/docs/applications/kegg_curation_implementation_status.md
@@ -91,7 +91,8 @@ by every stage to read/write its checkpoint file.
 
 ## `kegg_rest.py` — KEGG REST API Client
 
-Thin, direct wrapper around `https://rest.kegg.jp`, using `requests`
+Thin, direct wrapper around the [KEGG REST API](https://www.kegg.jp/kegg/rest/keggapi.html)
+(`https://rest.kegg.jp`), using `requests`
 directly rather than the `kegg_pull` PyPI package.
 
 **Why not `kegg_pull`**: `kegg_pull` 3.2.3 (latest on PyPI as of
diff --git a/docs/platforms/running_on_polaris.md b/docs/platforms/running_on_polaris.md
index 189c364..7a96591 100644
--- a/docs/platforms/running_on_polaris.md
+++ b/docs/platforms/running_on_polaris.md
@@ -210,8 +210,10 @@ Training Zone A: 100%|██████| 30/30 [00:31<00:00, 1.02s/it, loss=8.9
   Use anchored excludes (`--exclude='/data'`) when staging the repo
   to Polaris.
 - **Outbound network is via proxy.** Set
-  `HTTP_PROXY=http://proxy.alcf.anl.gov:3128` for HF downloads. The
-  launcher exports these by default; use `--no-proxy` for offline runs.
+  `HTTP_PROXY=http://proxy.alcf.anl.gov:3128` for HF downloads (an HTTP
+  proxy endpoint, not a browsable page — see the
+  [ALCF getting-started docs](https://docs.alcf.anl.gov/polaris/getting-started/)).
+  The launcher exports these by default; use `--no-proxy` for offline runs.
 - **Pre-cache HF models.** Compute nodes go through the proxy too,
   which is rate-limited and intermittently blocks. Download models
   on a login node first: