Add disclosure boundary and sample data

DISCLOSURE.md

@@ -0,0 +1,19 @@
+# Disclosure Boundary
+
+This repository shows an auditable AI-assisted alpha research loop without publishing the private parts of a real research book.
+
+## Included
+
+- Synthetic panel generation, formula parsing, candidate evaluation, and rejection accounting.
+- Tiny sample candidate records that show the public research-log schema.
+- Claim ceilings for what the evidence supports and does not support.
+
+## Excluded
+
+- Complete alpha libraries, live candidate pools, production thresholds, execution code, and capital-allocation logic.
+- Raw research logs, vendor data, credentials, account metadata, and any non-public dataset.
+- Any claim that the demo formulas are tradable.
+
+## Sample Data Rule
+
+Files under `sample_data/` are synthetic fixtures. They document format and validation behavior only.

README.md

@@ -32,7 +32,7 @@ python -m unittest discover -s tests -v
 
 ## Evidence boundary
 
-The first public record is intentionally conservative. An AutoAlpha-style study ran through the search and validation pipeline, but the frozen protocol produced zero final candidates. See [`evidence/validation-summary.md`](evidence/validation-summary.md), [`reports/rejection-gallery.md`](reports/rejection-gallery.md), and [`research_state.json`](research_state.json).
+The first public record is intentionally conservative. An AutoAlpha-style study ran through the search and validation pipeline, but the frozen protocol produced zero final candidates. See [`evidence/validation-summary.md`](evidence/validation-summary.md), [`reports/rejection-gallery.md`](reports/rejection-gallery.md), and [`research_state.json`](research_state.json). The public disclosure boundary is recorded in [`DISCLOSURE.md`](DISCLOSURE.md), with a tiny synthetic candidate-log fixture in `sample_data/`.
 
 ## Layout
 

sample_data/candidate_records_sample.jsonl

@@ -0,0 +1,3 @@
+{"formula":"rank(delta(amount, 3))","oos_rank_ic":0.012,"oos_net_return":-0.0011,"turnover":0.71,"promoted":false,"reason":"too_costly"}
+{"formula":"rank(mean(returns, 5))","oos_rank_ic":-0.004,"oos_net_return":-0.0007,"turnover":0.42,"promoted":false,"reason":"low_oos_rank_ic"}
+{"formula":"rank(delta(close, 2)) - rank(mean(returns, 5))","oos_rank_ic":0.018,"oos_net_return":-0.0003,"turnover":0.58,"promoted":false,"reason":"below_gate"}

tests/test_sample_data.py

@@ -0,0 +1,19 @@
+import json
+import unittest
+from pathlib import Path
+
+
+class SampleDataTests(unittest.TestCase):
+    def test_candidate_records_are_bounded_negative_examples(self):
+        records = [
+            json.loads(line)
+            for line in Path("sample_data/candidate_records_sample.jsonl").read_text(encoding="utf-8").splitlines()
+        ]
+
+        self.assertEqual(len(records), 3)
+        self.assertTrue(all(record["promoted"] is False for record in records))
+        self.assertTrue(all("formula" in record and "reason" in record for record in records))
+
+
+if __name__ == "__main__":
+    unittest.main()