Track record · updated 29 Oct 2026, 06:46 UTC
Hyperliquid whale signal track record
Every signal type DexiScan publishes, graded forward since 21 Jul 2026 (100 days). A type counts as working only when its whole 95% interval clears its benchmark. The types that have not cleared it are listed with the rest.
- History
- 100 days
- Beat a random entry
- 2 of 7 measured
- Beat the 4h benchmark
- 0 of 12 scored
- Market base rate, 4h
- 50.3% n = 16,714
01Edge over a random entry
One fixed exit rule replayed on every signal, compared with the same rule entered at random on the same coin and side. The result is measured in the coin's own volatility (σ), so a calm coin and a volatile one count alike. A family beats random only when its interval clears zero.
| Signal | Episodes | Days | Edge | 95% interval | Verdict |
|---|---|---|---|---|---|
Cross-signal cross_signal | 2,359 · 204 pending | 84 | +0.077σ | +0.015σ to +0.140σ | BEATS RANDOM |
Whale sentiment, 4h window sentiment_4h_v5 | 1,245 · 103 pending | 73 | +0.046σ | -0.011σ to +0.103σ | UNPROVEN |
Single-whale move whale_move | 1,131 · 234 pending | 84 | -0.022σ | -0.182σ to +0.138σ | UNPROVEN |
Whale sentiment, 12h window sentiment_12h_v5 | 969 · 92 pending | 72 | +0.054σ | -0.000σ to +0.107σ | UNPROVEN |
Whale sentiment, 24h window sentiment_24h_v5 | 742 · 80 pending | 72 | +0.042σ | -0.021σ to +0.104σ | UNPROVEN |
Accumulation, weak band, 4h window signal_accumulation_weak | 134 · 7 pending | 60 | +0.098σ | -0.014σ to +0.210σ | UNPROVEN |
Accumulation, weak band, 8h window signal_accumulation_8h_weak | 126 · 10 pending | 55 | +0.115σ | +0.011σ to +0.220σ | BEATS RANDOM |
Accumulation signal, 4h window signal_accumulation | 4 · 1 pending | 3 | -0.015σ | -0.644σ to +0.614σ | TOO THIN |
Accumulation signal, 8h window signal_accumulation_8h | 1 | 1 | +0.581σ | — | TOO THIN |
02Hit rate at 4 hours
The share of calls whose direction was right 4 hours later. A call is scored only when the move cleared 0.25 times that coin's own volatility; smaller moves are left out as noise. The benchmark is what the same calls would have scored from market drift alone, and lift is the difference.
| Signal | Calls | Hit rate | Benchmark | Lift | 95% interval | Verdict |
|---|---|---|---|---|---|---|
Cross-signal cross_signal | 3,000 | 51.0% | 49.9% | +1.1 pp | 48.4–53.6% | UNPROVEN |
Whale sentiment, 1h window sentiment_1h_v5 | 1,603 | 52.7% | 50.1% | +2.6 pp | 48.9–55.9% | UNPROVEN |
Whale sentiment, 4h window sentiment_4h_v5 | 1,338 | 51.5% | 50.0% | +1.5 pp | 48.4–54.4% | UNPROVEN |
Whale sentiment, 12h window sentiment_12h_v5 | 1,047 | 51.7% | 50.0% | +1.7 pp | 48.3–54.6% | UNPROVEN |
Whale sentiment, 24h window sentiment_24h_v5 | 814 | 50.4% | 49.9% | +0.4 pp | 46.6–53.8% | UNPROVEN |
Smart-money flow burst, 15m smart_flow_15m | 572 | 52.8% | 50.2% | +2.6 pp | 49.4–56.3% | UNPROVEN |
Divergence divergence | 462 | 50.6% | 49.9% | +0.7 pp | 46.3–54.9% | UNPROVEN |
Accumulation, weak band, 8h window signal_accumulation_8h_weak | 275 | 52.7% | 52.0% | +0.7 pp | 46.6–59.9% | UNPROVEN |
Accumulation, weak band, 4h window signal_accumulation_weak | 205 | 61.5% | 53.0% | +8.5 pp | 52.7–70.2% | UNPROVEN |
Accumulation signal, 4h window signal_accumulation | 105 | 65.7% | 61.3% | +4.4 pp | 51.8–82.1% | UNPROVEN |
Daily open-interest report oi_daily_report | 92 | 46.7% | 47.4% | -0.7 pp | 37.0–56.5% | UNPROVEN · thin |
Accumulation signal, 8h window signal_accumulation_8h | 31 | 38.7% | 51.0% | -12.3 pp | 13.3–62.9% | UNPROVEN · thin |
Too few calls to score yet, so no rate is quoted (under 30 calls or under 10 separate days): signal_accumulation_weak_retr (2 calls), signal_accumulation_retr (1 call).
03How to read the verdicts
- BEATS and BEATS RANDOM: the whole 95% interval sits above the benchmark, or above zero for the edge.
- UNPROVEN: the interval still crosses the benchmark. The headline number may be above it, but the data cannot yet tell it apart from luck.
- INVERTED and WORSE THAN RANDOM: the whole interval sits below.
- NO DATA and TOO THIN: too few calls or days to read anything.
The intervals are wide on purpose. Calls that overlap in time are not independent, so the bootstrap resamples whole days instead of single calls, and a verdict takes longer to arrive. Most signal types need weeks to months more data before their interval can clear the benchmark.
More on the method: why a hit rate needs a base rate, how a signal is graded, what an accumulation signal is.
04Data
The hit-rate table as CSV: track-record.csv. The interactive version adds the 1h, 12h and 24h horizons, splits by direction and confidence, and the upside and drawdown of each signal type.