Refresh dataset card and static metadata

README.md

@@ -4,11 +4,14 @@ pretty_name: Chainticks CFTC COT
 tags:
 - finance
 - crypto
+- data-analysis
 - parquet
 - duckdb
 - pandas
 - polars
 - agent-friendly
+task_categories:
+- tabular-regression
 configs:
 - config_name: default
   data_files:
@@ -20,6 +23,13 @@ configs:
 
 Normalized CFTC Commitments of Traders legacy futures rows from public-domain CFTC archives.
 
+## Current Scope
+
+- **Coverage:** CFTC legacy futures Commitments of Traders rows normalized from public-domain yearly archives.
+- **Best for:** Macro positioning features, futures crowding studies, and slower-moving cross-asset context.
+- **Cadence:** append-only partitions, normally refreshed by the Chainticks nightly publisher.
+- **Timestamps:** UTC ISO-8601 strings when available. Null timestamp fields mean the source event did not include a timestamp at collection time.
+
 ```python
 import pandas as pd
 
@@ -42,6 +52,10 @@ LATEST_DATE.txt
 
 Rows must have `source_kind` in `['public_domain']`. Chainticks does not publish venue REST/API resale data in this dataset.
 
+## Known Limitations
+
+COT data is weekly and delayed. It should be used as positioning context, not intraday execution data.
+
 ## Agent Prompt Snippet
 
 Use this dataset as append-only market context. Read `LATEST_DATE.txt`, inspect `_schema.json`, then load the relevant Parquet partition. Treat timestamps as UTC and preserve `source_kind` in downstream analysis.