Reassign pair-level 80/10/10 train-val-test splits stratified by culture

README.md

@@ -135,3 +135,7 @@ Please cite / credit the upstream resources this corpus builds on:
 Synthetic vignettes and translations were generated with OpenAI **gpt-4o-mini**.
 This is a **research pilot**, not a validated cultural benchmark. Norms are simplified
 etiquette descriptors and may not represent all speakers of a culture.
+
+## Splits
+
+Pair-level **80% / 10% / 10%** `train` / `val` / `test`, stratified by **culture**. Conforming and violating mates always share a split. The same content pair uses the same split across all languages (parallel multilingual holdout).

norms.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f6b1c7ec82c60a6d36ab275f720f8fc931951a831bbad8d0d40fe964b2e7161a
-size 30016
+oid sha256:7669b00790a90a570208e50399c6b6eda1e6f7f397842def9abc0c3e599cdb1a
+size 30064

samples.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e85f7a23cc20ed2c59356c74f6f3ba245ed770c3d87e5b8a6612378dade3bdcd
-size 5241876
+oid sha256:40646258b4e367b6ad51b1f8e23773a4a4f36c3ca6a03c1251ba7e079ab468b2
+size 5250161

seed_samples.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe75719c4e4f76d3729c05b8a8c9bf134d7299a0aa6c55fa6d6ffab4f8fd9f76
-size 57393
+oid sha256:4e644e9f7a4fc55c662a582210d9be53067e8f65a583e07094d71a4ad7778e11
+size 57449