card: replace sibling distractor numbers with this model's own measurements (unicode Tibetan)

README.md

@@ -102,20 +102,16 @@ print(q @ p.T)   # cosine similarities
 ## Evaluation
 
 R@1 on the mitra retrieval benchmark, **hardened setting: 25,000 in-language
-distractor passages added to every candidate pool** (each model in its native
-input convention). The mitra column reports Qwen3.5-Mitra-9B, the sibling
-checkpoint of this model (same base and training recipe; on gold pools the
-two differ by at most ~2 points per task):
+distractor passages added to every candidate pool** (this model's own
+measurements; Tibetan given in native script):
 
-| task | Qwen3.5-Mitra-9B (sibling)* | gemma-2-mitra-e | BGE-M3 |
+| task | this model | gemma-2-mitra-e | BGE-M3 |
 |---|---|---|---|
-| Multilingual parallels (sa/bo/zh) | **0.862** | 0.841 | 0.001 |
-| Multilingual asymmetric (sentence→window) | **0.789** | 0.581 | 0.000 |
-| English → classical | **0.900** | 0.879 | 0.068 |
-| Verse → commentary | **0.501** | 0.384 | 0.138 |
-| English question → passage | **0.522** | 0.473 | 0.020 |
-
-\* sibling checkpoint, not this exact model — see note above.
+| Multilingual parallels (sa/bo/zh) | **0.846** | 0.841 | 0.001 |
+| Multilingual asymmetric (sentence→window) | **0.788** | 0.581 | 0.000 |
+| English → classical | 0.876 | 0.879 | 0.068 |
+| Verse → commentary | **0.483** | 0.384 | 0.138 |
+| English question → passage | **0.523** | 0.473 | 0.020 |
 
 **This checkpoint's own results** on gold pools (per-subtask pools without
 added distractors):