diff --git a/docs/examples/t0-family.ipynb b/docs/examples/t0-family.ipynb
index 1afc8c3..dcc8c5d 100644
--- a/docs/examples/t0-family.ipynb
+++ b/docs/examples/t0-family.ipynb
@@ -28,7 +28,7 @@
     "| **t0-beta** | `theforecastingcompany/t0-beta` | ~256M | 21 (0.01\u20130.99) |\n",
     "| **t0-gamma** | `theforecastingcompany/t0-gamma` | ~256M | 21 (0.01\u20130.99) |\n",
     "\n",
-    "T0 is a decoder-style patch transformer that alternates attention over time with attention across variates. It forecasts zero-shot and natively accepts known-future covariates (see the [exogenous variables example](exogenous-variables.ipynb)); this notebook uses the univariate API.\n",
+    "T0 is a decoder-style patch transformer that alternates attention over time with attention across variates. It forecasts zero-shot and natively accepts known-future covariates; this notebook uses the univariate API.\n",
     "\n",
     "`T0` also exposes inference options from `tfc-t0`, all off by default and available for every checkpoint:\n",
     "\n",
diff --git a/docs/examples/timestral-family.ipynb b/docs/examples/timestral-family.ipynb
index 5e5d335..fc391e6 100644
--- a/docs/examples/timestral-family.ipynb
+++ b/docs/examples/timestral-family.ipynb
@@ -30,7 +30,7 @@
     "| **Timestral-123M** | `mistralai/Timestral-123M` | 123M |\n",
     "| **Timestral-209M** | `mistralai/Timestral-209M` | 209M |\n",
     "\n",
-    "Timestral forecasts zero-shot with 21 quantile levels (0.01\u20130.99), up to 32,768 steps of context, and native known-future covariates (see the [exogenous variables example](exogenous-variables.ipynb)). This notebook compares all four sizes on event pageview data using the univariate API.\n",
+    "Timestral forecasts zero-shot with 21 quantile levels (0.01\u20130.99), up to 32,768 steps of context, and native known-future covariates. This notebook compares all four sizes on event pageview data using the univariate API.\n",
     "\n",
     "By default, `Timestral` sets `symmetric_averaging=True` and `clip_nonnegative=True`, matching Mistral's published benchmark settings.\n",
     "\n",