Update README.md

README.md

@@ -49,21 +49,7 @@ Tasks are `kind: forecast` (labels shipped as-is) or `kind: autocomplete`. This
 **train-only** release: tasks ship a single `train.parquet` and `val_timestamp` /
 `test_timestamp` are null, designed for pretraining, not held-out evaluation.
 
-## RelBench decontamination
-
-Databases whose source data overlaps with RelBench benchmark datasets are excluded, so
-models pretrained on The Join do not see RelBench evaluation data. Exclusions are based
-on exact source match (same site, dump, or API): the McAuley Amazon reviews books subset
-(rel-amazon), AACT ClinicalTrials.gov (rel-trial), RateBeer reviews (rel-ratebeer),
-Formula 1 race history (rel-f1), and all copies of the stats.stackexchange.com dump
-(rel-stack), including those republished under other names by the CTU relational
-repository ("stats") and BIRD ("codebase_community"). The same rule is applied against
-the dbinfer family (4DBInfer): the Kaggle RetailRocket dump (dbinfer-retailrocket) and
-the Seznam.cz wallet data (dbinfer-seznam, also distributed by the CTU relational
-repository) are excluded; dbinfer's amazon (Books) and stackexchange
-(stats.stackexchange.com) sources coincide with the RelBench exclusions above. Datasets
-drawn from different subsets of a shared source family (e.g. other Amazon categories,
-other Stack Exchange sites) are retained.
+Databases whose source data overlaps with RelBench and dbinfer benchmark datasets are excluded.
 
 ## Licensing