Apply explicitly selected dataset corrections

README.md

@@ -361,6 +361,17 @@ configs:
   - split: test
     path:
     - data/test/STAR.parquet
+- config_name: StudyChat
+  data_files:
+  - split: train
+    path:
+    - data/train/StudyChat.parquet
+  - split: validation
+    path:
+    - data/validation/StudyChat.parquet
+  - split: test
+    path:
+    - data/test/StudyChat.parquet
 - config_name: Taskmaster-4
   data_files:
   - split: train
@@ -598,7 +609,7 @@ Roles are already oriented for user-simulator training:
 <!-- available-datasets:start -->
 ## Available Datasets
 
-**Dataset count: 50**
+**Dataset count: 51**
 
 Select an individual dataset by its name below, or use `default` for the combined collection.
 
@@ -616,6 +627,9 @@ Datasets are grouped by behavioral category:
     "AMI", "Humanual-Email", "ICSI", "MutualFriends", "deli-corpus", "fomc-corpus", "gap-corpus",
     "llm-facilitation-corpus", "spolin-corpus"
   ],
+  "Education & Tutoring": [
+    "StudyChat"
+  ],
   "General Conversation & Social Interaction": [
     "Humanual-Chat", "LMSYS-Chat-1M", "MSC", "Topical-Chat", "chitchat-dataset"
   ],

SOURCES.md

@@ -300,3 +300,50 @@ of interior shared passages, semantic duplicates or external benchmark items.
 Anonymization placeholders and original AI errors remain in context. Every
 exported message was compared with source; sampled semantic review cannot
 establish the quality or human authorship of every individual record.
+
+## StudyChat (OdysSim processed histories)
+
+Credit: Hunter McNichols, Fareya Ikram and Andrew Lan, *The StudyChat Dataset:
+Analyzing Student Dialogues With ChatGPT in an Artificial Intelligence Course*,
+LAK 2026. [Original dataset](https://huggingface.co/datasets/wmcnicho/StudyChat)
+and [paper](https://arxiv.org/abs/2503.07928).
+The original source documents consenting human students interacting with an AI
+assistant in UMass COMPSCI 383 and declares **CC BY 4.0**. Preserve the original
+source's terms; OdysSim's collection metadata does not relicense these records.
+No blanket license is assigned to this combined collection.
+
+Prepared source: [OdysSim split release](https://huggingface.co/datasets/Xuhui/sft_processed_large_split/tree/ee332b10508c0a6fbaa86d5de729c89d34f68964/studychat),
+revision `ee332b10508c0a6fbaa86d5de729c89d34f68964`. Original-source verification used revision
+`24d7987d9fbb30d9da12acc53455a10f1cdd2d7f`. All 2,214 supplied histories match an original snapshot
+after comparison-only outer whitespace trimming and omission of blank messages.
+Original interaction rows and complete conversation histories have different
+counts; no original-source interaction rows were additionally imported.
+
+Human student `user` messages become training-target `assistant` messages;
+AI tutor `assistant` messages become `user` context. Supplied system personas,
+retained text, order, message boundaries, consecutive speakers, and final
+student messages remain unchanged. No personas or dialogue are generated.
+The 11 tutor-first histories were verified against original history snapshots
+and retained without reordering. Source-recorded order is not an independent
+verification of real-world chronology. References to local code, files, or
+assignments and coherent requests for clarification remain in the dialogue;
+no unavailable material was fabricated or appended.
+
+Excluded 478 histories containing only one student message (440 train, 10
+validation, 28 test), and six shorter training prefixes in favor of their
+longer conversations. Retained personas are each longer record's own context.
+Counts: **1,523 train; 20 validation; 187 test; 1,730 total**.
+OdysSim assignments are unchanged, with `val` renamed to `validation`.
+
+Supplied personas are inferred context and may describe later dialogue. They
+are not verified prior student instructions. The splits share student IDs and
+are **not student-disjoint or verified leakage-free evaluations**. Original
+AI errors remain as context; this is a human-behavior training source rather
+than a set of verified tutoring answers.
+
+Full normalized dialogues and substantive full-prefix overlaps were checked
+within the retained source and against the published collection. This does
+not establish the absence of arbitrary semantic or interior-subsequence overlap.
+Every exported message was compared with its OdysSim source. Seeded samples,
+all 11 tutor-first histories, and all six prefix pairs were reviewed; sampled
+semantic checks do not establish the quality of every individual utterance.

data/test/StudyChat.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51588f34ed1bb7fb90014b88114e0ff43f6ad9bd114c418476c39439f5f1d5c0
+size 1471339

data/train/StudyChat.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98701683070e6d819132c28429aa6666e901c80bd9af610986563dc09ce5f0dc
+size 13375122

data/validation/StudyChat.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9feafdd613c617701207a4c27be420c990300bef807e64b984b811174d4f7662
+size 165574