AutoScientist Part 2 source dataset

README.md

@@ -23,9 +23,9 @@ Built for the [Adaption Labs AutoScientist Challenge](https://adaptionlabs.ai/bl
 
 | | |
 |---|---|
-| Rows | **3,264** (2,896 train / 368 eval) |
-| Task families | **12** |
-| Companies covered | **2,157** of 4,230 available |
+| Rows | **3,318** (2,939 train / 379 eval) |
+| Task families | **13** |
+| Companies covered | **2,159** of 4,230 available |
 | Response length | median **57** words (p95 91, max 96) |
 | Duplicate instructions | **0** |
 | Source | SEC EDGAR XBRL company facts |
@@ -41,6 +41,7 @@ Built for the [Adaption Labs AutoScientist Challenge](https://adaptionlabs.ai/bl
 | `peer_comparison` | 350 |
 | `risk_flag` | 350 |
 | `earnings_headline` | 350 |
+| `premise_check` | 54 |
 | `filing_navigation` | 54 |
 | `metric_definition` | 36 |
 | `data_caveat` | 24 |
@@ -53,18 +54,18 @@ challenge trained on ~780-word CoT and *lost* to its own base model on judged
 win-rate — twice. Length is not quality, and on this metric it is actively
 harmful.
 
-**Split by company, not by row.** The 368 evaluation rows come from
+**Split by company, not by row.** The 379 evaluation rows come from
 507 companies that appear **nowhere** in training
 (verified: 0 overlapping companies). A row-level split would let the
 same company appear on both sides and report memorisation as generalisation.
 
-**Breadth over depth.** 12 task families rather than one schema repeated.
+**Breadth over depth.** 13 task families rather than one schema repeated.
 The target is competence across the domain, including tasks phrased in ways this
 dataset does not contain.
 
 ## Verification
 
-**1,534 numeric claims across both datasets were independently re-derived from source and every one matches the figure stated in the response.**
+**1,660 numeric claims across both datasets were independently re-derived from source and every one matches the figure stated in the response.**
 
 The verifier (`verify.py`, included in this repo) does not import the generator's
 arithmetic. It parses

profile.json

@@ -1,9 +1,10 @@
 {
-  "total_rows": 3264,
-  "train_rows": 2896,
-  "eval_rows": 368,
-  "task_families": 12,
+  "total_rows": 3318,
+  "train_rows": 2939,
+  "eval_rows": 379,
+  "task_families": 13,
   "by_family": {
+    "premise_check": 54,
     "fundamental_lookup": 350,
     "yoy_growth": 350,
     "margin_analysis": 350,
@@ -17,7 +18,7 @@
     "filing_navigation": 54,
     "data_caveat": 24
   },
-  "companies_covered": 2157,
+  "companies_covered": 2159,
   "companies_available": 4230,
   "held_out_companies": 507,
   "response_words": {
@@ -25,7 +26,7 @@
     "p50": 57,
     "p95": 91,
     "max": 96,
-    "mean": 57.9
+    "mean": 57.7
   },
   "duplicate_instructions": 0,
   "split_leakage_ciks": 0,