diff --git a/docs/assets/lora-validation/qwen3-5-9b-dpo-hhh.png b/docs/assets/lora-validation/qwen3-5-9b-dpo-hhh.png
new file mode 100644
index 0000000..39f47e7
Binary files /dev/null and b/docs/assets/lora-validation/qwen3-5-9b-dpo-hhh.png differ
diff --git a/docs/lora_validation.md b/docs/lora_validation.md
index 12cde6d..c7302b0 100644
--- a/docs/lora_validation.md
+++ b/docs/lora_validation.md
@@ -122,8 +122,13 @@ python -m tinker_cookbook.recipes.preference.dpo.train \
 | chosen reward | −0.0001 | 0.0855 |
 | rejected reward | 0.0011 | 0.0353 |
 
+![DPO loss per step on Spindle and Tinker](assets/lora-validation/qwen3-5-9b-dpo-hhh.png)
+
 - Margin and accuracy rise over the run, with chosen rewards pulling away from
   rejected ones.
+- The same command on Tinker sees identical batches and follows the same loss
+  trajectory (step 9 dpo_loss: 0.6795 on Spindle, 0.6837 on Tinker).
 - A warm step takes about 21 s, plus 9–14 s to compute reference logprobs on
   the sampler pool.
-- [W&B run](https://wandb.ai/modal-labs/spindle-dpo-validation/runs/0z0obns5).
+- W&B: [Spindle](https://wandb.ai/modal-labs/spindle-dpo-validation/runs/0z0obns5),
+  [Tinker](https://wandb.ai/modal-labs/spindle-dpo-validation/runs/3zbxmlbv).