English · 한국어
RESEARCH 001–005 · SYNTHESIS
If you remember only three things
- A trading statistic is meaningless without its measurement rules. Timeframe, horizon, event timing, and data alignment changed the interpretation materially.
- Several simple market narratives did not survive a frozen baseline test. A basic liquidity-sweep rule and a basic funding × OI grid did not produce robust short-horizon directional edges.
- Negative results are useful. They narrow the search space and stop weak assumptions from being repeated as facts.
What we confirmed
1. FVG revisit rates depend heavily on the clock
Research 001 showed that a BTC 15m FVG’s revisit percentage rises as the observation window gets longer. Research 002 then showed that the apparent timeframe effect changes depending on whether we hold bar count or elapsed time constant.
Under equal bar counts, 1m, 5m, and 15m first-touch rates were very close. Under the same 60-minute clock window they separated sharply: 91.41%, 80.98%, and 67.45%.
2. Time-of-day association exists, but causality is still open
Research 003 found a higher 20-bar first-touch rate for New York-labeled FVGs, while the same window had a lower event frequency per observed hour. The association survived basic direction and volatility checks, but the study does not isolate the causal mechanism.
What we did not confirm
3. “Liquidity sweep = immediate reversal” did not hold as a standalone rule
Research 004 measured 5,071 confirmed-pivot sweep events. Five bars later, the reversal-direction return was positive 50.48% of the time and the mean was -1.27 bps. More than half of events later closed beyond the reference level again.
4. Funding × OI did not separate short-horizon direction cleanly
Research 005 classified 1,785 Bybit BTCUSDT funding settlements into nine Funding × OI-change states. Every four-hour cell confidence interval crossed zero, and none of the 36 pairwise comparisons remained significant after FDR correction.
The clearer descriptive pattern was volatility: large relative OI changes coincided with larger realized movement, not a reliable next-direction signal.
What we still do not know
- Whether pre-specified conditional filters can turn the sweep baseline into a useful event model.
- Whether the FVG and session findings replicate on other assets, venues, or future out-of-sample periods.
- Whether any measured event survives fees, slippage, stop/target ordering, and position sizing as an executable strategy.
- Whether a more carefully designed funding/OI model predicts volatility or risk better than direction.
The five studies in one line each
001 · FVG revisit — Longer observation windows produce higher measured revisit rates.
002 · Timeframe — Equal bars look similar; equal clock time does not.
003 · Time window — New York-labeled events revisited more often in this sample, but causality remains unresolved.
004 · Liquidity sweep — The simple public sweep rule did not show a reliable short-horizon reversal edge.
005 · Funding × OI — Direction did not separate robustly; OI-change states showed a clearer volatility difference.
How to use these studies
Use them as measurement baselines, not as ready-made signals. A result is most useful when it tells us which assumptions deserve another frozen test—and which assumptions should stop being treated as obvious.
Technical registry & reproducibility
The separate data index lists each study’s venue, sample, event count, primary outcome, code version, SHA-256 anchors, and robustness checks. Reproducibility packages are retained for Research 001–005.