How Research works
How EdgeDepth turns a market move into a counted answer on recorded Binance microstructure. Hyperliquid historical and cross-exchange research are not released yet.
Four steps
- Find a move. Pick one in Explore or Investigate a move. The readings that stood out before it are a lead, not evidence: the move was chosen afterwards.
- Test it. The Workbench counts every match and what followed. Every rate sits beside its count; missing values count separately, never as zero.
- Plan a trade. Add an entry, stop, target or trailing exit to see what each trade made after costs. The win rate sits beside the rate needed to break even.
- Watch it. Saved keeps results and alerts. An alert means the conditions recurred, not a trade signal. Pause stops it.
Habits that keep you from fooling yourself
A convincing chart is the easiest thing to be wrong about. These habits catch most of it:
- Write the question down first, precisely enough to count, before you look at what followed. Nothing runs or uses allowance until you press Run.
- Compare with something. A rate means little without the reference beside it.
- Mind the order of events. “Price got back up eventually” and “price got back up before it fell further” are different questions, and only the second matches a trade with a stop.
- Check another period, and make sure one day or one market is not carrying the result.
- Watch it forward with Saved before you trust it.
Most beliefs tested this way turn out to be noise. Finding that out cheaply is the point.
What a result does not tell you
- Not a trade result. How often price moved is not what a trade made. Plan adds the entry, exits and costs; neither is a forecast.
- Not a cause. The reference is every minute of those markets and dates, matches included: not a control group. A gap may be chance or repeated tries.
- Not independent. Nearby minutes share what happened next, and markets often move together. The match count overstates the evidence.
- Not typical. A replay picked for a big later move is one case. Open matches that went the other way too.
Keep reading
- Research limits: what a result cannot show.
- Base rates and outcome distributions: comparing with the reference.
- Selection bias: picked moves and repeated tests.
- Reproducibility: why new data can change results.
- More: research demo, coverage, query grammar.
Looking for a calculation? Open the Reading library. For recorded exercises, browse guided lessons.