Hello,

I am looking for research and practical validation methods for reducing false trend-reversal alerts without materially reducing detection or delaying the signal. The intended evaluation is separate for 30-minute, 1-hour, 2-hour and daily charts, with one frozen configuration per timeframe and decisions at fixed intrabar checkpoints. The objective is detection quality and latency, not profitability optimization.

Which causal features would you prioritize: static bid/ask depth imbalance, event-based order-flow imbalance, absorption, or price-path features available from trades/OHLCV? In particular:

1. Which primary studies or reproducible examples show useful prediction beyond the next tick, rather than only immediate price impact?
2. What minimum data and event sequencing are needed to distinguish these mechanisms reliably? Are top-of-book snapshots sufficient for any of them, and which require timestamped updates, cancellations or aggressor-side trades?
3. How would you design a chronological test that measures the trade-off between false alerts, missed reversals and signal delay without future-bar information leaking into the features?

I found the recent discussion on Tick.SaleCondition and historical/live differences (https://www.quantconnect.com/forum/discussion/21494/is-tick-salecondition-populated-for-us-futures-in-live-trading-aggressor-side/). My question is about feature selection and validation methodology, not a guarantee that a particular feed provides these fields.

References to primary papers or small reproducible examples would be especially helpful. Thank you.