Apply explicitly selected dataset corrections

README.md

@@ -97,6 +97,17 @@ configs:
   - split: test
     path:
     - data/test/CraigslistBargains.parquet
+- config_name: Deal or No Deal
+  data_files:
+  - split: train
+    path:
+    - data/train/Deal or No Deal.parquet
+  - split: validation
+    path:
+    - data/validation/Deal or No Deal.parquet
+  - split: test
+    path:
+    - data/test/Deal or No Deal.parquet
 - config_name: DuRecDial-2.0
   data_files:
   - split: train
@@ -477,7 +488,7 @@ Roles are already oriented for user-simulator training:
 <!-- available-datasets:start -->
 ## Available Datasets
 
-**Dataset count: 39**
+**Dataset count: 40**
 
 Select an individual dataset by its name below, or use `default` for the combined collection.
 
@@ -502,7 +513,8 @@ Datasets are grouped by behavioral category:
     "news-interview", "wizard_of_internet", "wizard_of_wikipedia"
   ],
   "Negotiation, Persuasion & Debate": [
-    "CaSiNo", "CraigslistBargains", "contextual-abuse", "diplomacy-corpus", "iq2-corpus"
+    "CaSiNo", "CraigslistBargains", "Deal or No Deal", "contextual-abuse", "diplomacy-corpus",
+    "iq2-corpus"
   ],
   "Preferences & Recommendations": [
     "CCPE-M", "DuRecDial-2.0", "INSPIRED", "OpenDialKG", "Redial"

SOURCES.md

@@ -95,3 +95,23 @@ context. Speech wording, boundaries and order are unchanged. New approximately
 training-only release. No generated personas or action paraphrases are added.
 Some retained speech refers to removed selection events; this conversion does not
 claim every exchange is self-contained without the original interface.
+
+## Deal or No Deal
+
+Original release: [Facebook AI Research, end-to-end-negotiator](https://github.com/facebookresearch/end-to-end-negotiator/tree/bbb93bbf00f69fced75d5c0d22e855bda07c9b78).
+The source LICENSE is [CC BY-NC 4.0](https://github.com/facebookresearch/end-to-end-negotiator/blob/bbb93bbf00f69fced75d5c0d22e855bda07c9b78/LICENSE).
+These source-specific terms apply to the reformatted files and do not change
+licenses of other datasets in this collection.
+
+Changes: preserve the existing participant perspectives once each; target speech
+becomes assistant, other speech becomes user. Formatting markers and selection
+actions are removed. The target's own item quantities and per-item utilities
+are formatted as system context; opponent utilities and final allocations are
+not included. Released speech strings and official splits are preserved, with
+val named validation. Single-utterance records are excluded; valid unmatched
+perspectives and failed negotiations remain. No missing counterpart is generated.
+
+The downloaded release has 12,234 perspective records. Removing 64 single-utterance
+records leaves 12,170, including 338 unmatched perspectives. Counterpart groups
+are inferred from ordered speech and aligned goals; original dialogue IDs are
+not supplied, so exported IDs identify native split files and line numbers.

data/test/Deal or No Deal.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a51374f67941b3b2602ce3c46a191b8b9fbf47277528d75b2f07bb7003b734c1
+size 74342

data/train/Deal or No Deal.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f6486935319c0432c8cc77315d10c5b2c97e84eb87ca9e17faac7abc2dc7778
+size 700311

data/validation/Deal or No Deal.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f02b5741a5b534b99744c9c12745209472781877432add738efa0d7de0b28fc
+size 77159