Initialize perp open data dataset

README.md

@@ -8,6 +8,10 @@ tags:
 - trading
 - parquet
 - time-series
+- pandas
+- duckdb
+- polars
+- mlcroissant
 task_categories:
 - tabular-regression
 ---
@@ -16,6 +20,16 @@ task_categories:
 
 Free, daily-updated perpetuals market data intended for quant research, backtesting, and market microstructure analysis.
 
+```python
+import pandas as pd
+
+DATE = "YYYY-MM-DD"
+URL = "https://huggingface.co/datasets/Chainticks/perp-data/resolve/main/hyperliquid_chain/trades/date={DATE}/part-0000.parquet"
+
+trades = pd.read_parquet(URL.format(DATE=DATE))
+print(trades.head())
+```
+
 This repository is initialized for **chain-derived perp DEX data**, starting with Hyperliquid. The public dataset must only contain records whose provenance is public chain/archive state, not venue REST API resale. The first production feed publishes Hyperliquid funding, trades, markets, open interest, and liquidations as partitioned Parquet under an explicit `hyperliquid_chain/` provider partition.
 
 ## Status
@@ -41,11 +55,30 @@ _manifest.json
 import pandas as pd
 
 repo = "Chainticks/perp-data"
-url = f"https://huggingface.co/datasets/{repo}/resolve/main/hyperliquid_chain/funding/date=YYYY-MM-DD/part-0000.parquet"
+date = "YYYY-MM-DD"
+url = f"https://huggingface.co/datasets/{repo}/resolve/main/hyperliquid_chain/funding/date={date}/part-0000.parquet"
 funding = pd.read_parquet(url)
 print(funding.head())
 ```
 
+```python
+import duckdb
+
+date = "YYYY-MM-DD"
+url = f"https://huggingface.co/datasets/Chainticks/perp-data/resolve/main/hyperliquid_chain/liquidations/date={date}/part-0000.parquet"
+rows = duckdb.sql("select symbol, count(*) as n from read_parquet(?) group by 1 order by 2 desc", [url]).df()
+print(rows)
+```
+
+```python
+import polars as pl
+
+date = "YYYY-MM-DD"
+url = f"https://huggingface.co/datasets/Chainticks/perp-data/resolve/main/hyperliquid_chain/open_interest/date={date}/part-0000.parquet"
+oi = pl.read_parquet(url)
+print(oi.head())
+```
+
 ## Provenance
 
 Eligible public rows use one of these `source_kind` values:
@@ -56,4 +89,22 @@ Eligible public rows use one of these `source_kind` values:
 
 API-sourced internal research rows are intentionally excluded from this public dataset.
 
+## Agent Prompt Snippet
+
+```text
+You can query Chainticks Perp Data directly from Hugging Face as partitioned Parquet. Use URLs shaped like:
+https://huggingface.co/datasets/Chainticks/perp-data/resolve/main/<provider>/<dataset>/date=YYYY-MM-DD/part-0000.parquet
+
+Valid provider for v1: hyperliquid_chain.
+Valid datasets: funding, trades, markets, open_interest, liquidations.
+Read _schema.json before generating queries. Read _manifest.json for available files, row counts, and UTC time ranges.
+Only treat rows as public-source eligible when source_kind is one of: on_chain_event, chain_rpc, hypercore_s3.
+```
+
+## Machine Metadata
+
+- Schema sidecar: `_schema.json`
+- Manifest sidecar: `_manifest.json`
+- Croissant metadata: `https://huggingface.co/api/datasets/Chainticks/perp-data/croissant`
+
 Chainticks is independent and is not affiliated with, endorsed by, or sponsored by Hyperliquid Labs or any protocol whose data appears here. Protocol names are used descriptively.

_manifest.json

@@ -1,6 +1,6 @@
 {
   "files": [],
-  "generated_at": "2026-05-10T11:58:07.573791+00:00",
+  "generated_at": "2026-05-10T12:02:42.040827+00:00",
   "public_source_kinds": [
     "chain_rpc",
     "hypercore_s3",

_schema.json

@@ -62,5 +62,6 @@
     ]
   },
   "public_dataset_rule": "Only chain-derived records are eligible for public parquet publishing.",
-  "schema_version": 1
+  "schema_version": 1,
+  "timestamp_policy": "All timestamps are UTC ISO-8601 strings. Missing values are encoded as Arrow nulls, not NaN sentinels."
 }