CID99 commited on
Commit
38bae05
·
verified ·
1 Parent(s): 9a4b7b2

Upload folder using huggingface_hub

Browse files
Files changed (2) hide show
  1. README.md +16 -0
  2. gold/test.parquet +2 -2
README.md CHANGED
@@ -97,6 +97,22 @@ Same fields as above, plus:
97
  |---|---|---|
98
  | `csrd_category` | string | ESRS category, or `none` (12 possible values) |
99
  | `esrs_subcategory` | string | ESRS subcategory, where applicable |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
100
 
101
  ### Regulatory regimes (`epoch`)
102
 
 
97
  |---|---|---|
98
  | `csrd_category` | string | ESRS category, or `none` (12 possible values) |
99
  | `esrs_subcategory` | string | ESRS subcategory, where applicable |
100
+ | `chain_of_thought` | string | **Expert's written reasoning** behind the label (see below) |
101
+
102
+ #### Expert reasoning (`chain_of_thought`)
103
+
104
+ Each annotation carries the annotator's written justification, structured in three steps:
105
+ (1) nature of the disclosed information, (2) materiality analysis under the CSRD double
106
+ materiality framework, (3) assessment of market surprise. Reasonings average roughly 620
107
+ characters.
108
+
109
+ This field is released because expert-written justifications on regulatory text are scarce.
110
+ It supports uses the labels alone do not: auditing *why* a paragraph was classified a given
111
+ way, studying where human and model reasoning diverge, chain-of-thought distillation, or
112
+ retraining an annotator on the same protocol.
113
+
114
+ It is a **single annotator's** reasoning, with no inter-annotator agreement measure — treat it
115
+ as documented judgment, not ground truth.
116
 
117
  ### Regulatory regimes (`epoch`)
118
 
gold/test.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9d9e6d01da95f8048a61cf9d4bd9a9c45cdf4417714c0d15663bb21a26bb3e4a
3
- size 81738
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86e933c5700365593851723f46d1307b1797a4c2936717fd74f65e00b2c468da
3
+ size 98377