Research is a work in progress. Estimated release: 1 December 2026. Our goal is to make it a search engine for trading setups in recorded crypto markets.
Binance USDT-MHEMIUSDT rose 82% on 27 August 2026
Picked after the rise, on one market.
Next stepWhat happened
Price opened the 05:00 hour at 0.007657 and reached a highest trade of 0.01395 in the 15:00 hour; the tested readings were taken at 04:59.
What stood out at 04:59
- Order book tilt0.9501Tested: 0.95 or higher
- Liquidation bunching0.6047Tested: 0.6 or higher
- 15-minute price change0.01379Tested: 0.013 or higher
04:59 always meets its own rules.
What the record says
30 minutes later, 21 of 44 were at or above the match price and 23 below; no reference was saved
Picked after the rise, on one market.
Evidence
Does it hold on other dates?
Nothing runs until you press Run.
Exactly what was counted3 conditions · 44 of 43,201 minutes · saved text
- Order book tilt 0.95 or higher. Size at the best bid against the best ask; above 0 means more bids, below 0 more asks.Book imbalance · top levels ≥ 0.95 · feature.book_imbalance
- Liquidation bunching 0.6 or higher. How bunched in time recent liquidations were; higher means more bunched, and it stays below 1.Cascade clustering ≥ 0.6 · feature.cascade_clustering_index
- 15-minute price change 0.013 or higher. Close to close change over the last 15 minutes, as a fraction of price.Return · 15m ≥ 0.013 · feature.ret_15m
Read at 1 minute before the move.
Read at 04:59 UTC on 27 August, one minute before the selected 05:00 start.
HEMI only, 4 August to 2 September 2026 (UTC, ending 3 September 00:00); all 3 readings had values on 43,201 of those minutes.
Reference: every HEMI minute over the same dates, matches included; it is not a matched control group.
Outcome: the close exactly 30 minutes later, not the highest or lowest price along the way.
The 05:00 start was picked automatically after the move.
HEMIUSDT stored readings end 2026-09-03 00:00 UTC (exclusive). Prices and tick replay have separate coverage.
The feature manifest spans the lead-up. Individual minutes or inputs can still be unavailable.
This is not a similarity search. Each count matches the exact readings above and nothing else.
Picked in hindsight: the counts include every other time these readings appeared, not just this move.
Fires often arrive as clusters of consecutive minutes, and forward windows overlap. A cluster is one episode, not many independent trials.
If these counts read weaker than the move suggested, that is the tool working: the lead-up of a picked move always looks special until it is counted everywhere else.
Saved study q#845ac2bb @ dataset e92709f8, counted 4 Sep 2026 07:57 UTC.
Saved summary text, as written 4 Sep 2026
Written automatically when this study was saved and kept as written; where a line differs from the counts above, the counts govern.
Minutes like this appeared 44 times in the 43,201 minutes searched (0.1%). What followed was close to a coin flip: 21 rose and 23 fell over the next 30 minutes (44 of the 44 matches have a complete 30 minutes forward).
After 1 hour this setup fell 2% or more in 14 of 44 complete windows (31.8%), against 13.5% with no condition (5,825 of 43,123): 2.36x.
This reports the time frame with the biggest gap from any minute, chosen after counting. Longer windows can run through the rise.
Best point inside 1 hour reached +5% or more in 12 of 44 complete windows; worst point fell 5% or more in 11 of 44.
About 1.4 a day, on 16 of 31 days: spread through the record rather than one event.
Picked after the rise, on one market.
When the matches happened
Of the 44 matches, 29 came before the rise began, 6 during it and 9 after it.
Neighbouring minutes overlap, so these are not 44 separate trials.
Picked after the rise, on one market.
Other time frames
The saved summary picked, after counting, the time frame with the biggest gap from any minute: 1 hour.
After 1 hour this setup fell 2% or more in 14 of 44 complete windows (31.8%), against 13.5% with no condition (5,825 of 43,123): 2.36x.
Along the way (the highest and lowest price inside 1 hour, not the close):
Best point inside 1 hour reached +5% or more in 12 of 44 complete windows; worst point fell 5% or more in 11 of 44.
Picked after the rise, on one market.
Readings at five times before the rise24 hours to 1 minute before
Highlights follow fixed rules, not significance tests. Proposed conditions round values or use ranges; inspect them before running.
24 hours before
No readings were recorded for this minute.
4 hours before
No readings were recorded for this minute.
1 hour before
No readings were recorded for this minute.
15 minutes before
No readings were recorded for this minute.
1 minute before · 27 Aug 04:59 UTC · 4 of 53 stood out
Order book tilt 0.95 or higherLiquidation bunching 0.6 or higher15-minute price change 0.013 or higher1-hour price change 0.022 or higherShow the full readings
Reading Value at that minute Stored rule Book imbalance · top levels feature.book_imbalance 0.9501 ≥ 0.95 Cascade clustering feature.cascade_clustering_index 0.6047 ≥ 0.6 Return · 15m feature.ret_15m 0.01379 ≥ 0.013 Return · 1h feature.ret_1h 0.0226 ≥ 0.022
Hourly candles around the risefull context, lead-up, move
The 05:00 start was picked after the move, and the hour before was already rising.
Hourly candles, UTC. The marked lines are the minutes the readings came from: 1 minute, 15 minutes, 1 hour, 4 hours and 24 hours before the selected start.
Other ways to test itsame dates
Same 3 conditions on HEMI, 5 July to 3 August 2026.
Each opens Review the test.
Running needs a free account and uses your allowance; you see the cost first.