Organization card

README.md

@@ -9,9 +9,12 @@ pinned: false
 
 # QuantFeatures AI
 
-**Institutional-quality, point-in-time crypto market features from tick data, for independent quants and small teams.**
+**Institutional-quality crypto market features from tick data, for independent quants and small teams: nearly 7 years of history, built for backtesting and live trading alike.**
 
-Large funds have data teams that turn raw tick data into clean research features. QuantFeatures provides that feature data without a data-engineering team: trade flow, trade-size statistics, covariance, liquidations, funding, open interest, and top of book.
+Large funds have data teams that turn raw tick data into clean, systematically prepared research features. QuantFeatures provides that feature data without a data-engineering team: trade flow, trade-size statistics, covariance, liquidations, funding, open interest, and top of book.
+
+- **Long, consistent history:** December 2019 to today, cleaned and prepared the same way across venues and years.
+- **Backtest to production:** historical and real-time features come from the same computation, so what you backtest is what you run live.
 
 ## Free samples
 
@@ -24,8 +27,7 @@ Hourly features for the first UTC day of each month, December 2019 to October 20
 | [bybit-hourly-features](https://huggingface.co/datasets/QuantFeatures/bybit-hourly-features) | Perpetual (USDT) | 17 |
 | [hyperliquid-hourly-features](https://huggingface.co/datasets/QuantFeatures/hyperliquid-hourly-features) | Perpetual (USDC) | 15 |
 
-- **Point-in-time:** each row covers one instrument over a half-open UTC interval `[start, end)`; use `end_time_utc` as the availability time to avoid look-ahead bias.
-- **Explicit missing values:** NaN means not computable, never a silent zero.
+- **Cleaned and documented:** exchange outages and missing data are handled the same way everywhere; NaN means not computable, never a silent zero.
 - **Easy to load:** Parquet, one config per module, one split per market.
 
 ```python
@@ -37,8 +39,8 @@ Field definitions, timestamp semantics, and coverage: [Feature Catalog](https://
 
 ## Built for AI
 
-- **Model-ready inputs:** point-in-time features with no look-ahead, explicit missing values, and standard Parquet that loads with `datasets` or pandas.
-- **Agent-ready:** an MCP (Model Context Protocol) server lets AI assistants and agents discover features by meaning, read exact definitions, check coverage, and fetch point-in-time rows. Available to customers on request.
+- **Model-ready inputs:** consistent features across venues and years, explicit missing values, and standard Parquet that loads with `datasets` or pandas.
+- **Agent-ready:** an MCP (Model Context Protocol) server lets AI assistants and agents discover features by meaning, read exact definitions, check coverage, and fetch rows as of any time. Available to customers on request.
 
 ## Full data on request