Clarify FineWeb comparison conclusion

README.md

@@ -101,7 +101,7 @@ For a URL-level comparison on `CC-MAIN-2025-26`, the deduplicated Ultra-FineWeb-
 | Recall against FineWeb | 94.30% | **90.61%** |
 | Precision | 70.69% | **90.16%** |
 
-Overall, this validation run shows that MinHash deduplication trades some recall for a substantially closer URL-set match to FineWeb: precision increases from 70.69% to 90.16%, while recall decreases from 94.30% to 90.61%. These figures describe URL overlap for this dump only; they do not imply that the two datasets contain identical documents or that the same metrics apply to all released dumps.
+On this validation dump, MinHash deduplication removes many URLs that are not present in FineWeb: precision increases from 70.69% to 90.16%, while recall decreases from 94.30% to 90.61%. This reflects a trade-off in URL coverage and deduplication, not a direct measurement of overall text quality. The figures describe URL overlap for this dump only; they do not imply that the two datasets contain identical documents or that the same metrics apply to all released dumps.
 
 ## 📁 Dataset Structure