Public register
This page exists so that nothing about the Oracle has to be taken on trust. It carries the headline figure, how the signal is built, the part of the data the figure does not cover, and the full price-by-outcome table — weak bands included.
- Hit rate
- 83.4%
- Markets covered
- BTC · ETH · SOL · XRP
- Slot length
- 5 min
on 136,252 verified slots
136,252 resolved slots across the four
One directional call per market, per slot
All figures on this page come from a single database snapshot taken at October 3, 2026 at 11:55 AM. Taking them at different moments would stop the totals adding up.
How the signal is built
The Oracle is a two-stage signal on a five-minute slot. A first read is written 11 seconds in, from momentum and order-flow features. At T+180s, with two minutes left to run, that read is confirmed against where the market has actually gone — and the confirmed signal is the product.
The number reflects exactly that: on 136,252 slots with a known outcome, the confirmed direction matched the result 83.4% of the time. We describe the confirmation step openly rather than presenting a single figure without its mechanism, because the mechanism is the reason the figure holds.
What the confirmation step is worth
The signal is emitted in two stages: a first read when the slot opens, then a confirmation at T+180s against how the market has actually moved. Only the confirmed read is sold. Splitting the verified slots by that step shows what it is worth — and what the discarded half looks like without it.
| Signal | Slots | Hit rate |
|---|---|---|
| Confirmed at T+180s — the product | 129,061 | 85.7% |
| Never confirmed — discarded, not sold | 6,843 | 41.9% |
The second row is the control, not an offer: it is what the first read scores when the market never moved enough to confirm it, which is why confirmation is the product and not the initial read. A register carrying only the confirmed half would be worth less than one that carries both. A further 348 verified slots had no move sampled at all and sit outside this split; at that sample size the rate is not worth reading, and it is printed so the rows reconcile.
What the hit rate does not cover
A slot is unverifiable when the resolution service could not retrieve its closing price. Those slots are not wrong calls — they are unknown, and they are excluded from the rate rather than counted as failures. The share moved from 0% in March to 73.8% in May, which is why it appears next to the rate and not in a footnote: without it, one month cannot be compared with another.
| Month | Emitted | No outcome | Share | Verified | Hit rate | |
|---|---|---|---|---|---|---|
| Mar 2026 | 23,548 | 0 | 0.0% | 23,548 | 76.8% | |
| Apr 2026 | 34,556 | 23,446 | 67.9% | 11,110 | 81.3% | |
| May 2026 | 35,710 | 26,341 | 73.8% | 9,369 | 87.4% | |
| Jun 2026 | 34,559 | 15,832 | 45.8% | 18,727 | 84.8% | |
| Jul 2026 | 35,712 | 15,445 | 43.3% | 20,267 | 83.6% | |
| Aug 2026 | 35,708 | 10,449 | 29.3% | 25,259 | 85.1% | |
| Sep 2026 | 34,558 | 8,953 | 25.9% | 25,605 | 85.7% | |
| Oct 2026 | 2,880 | 513 | 17.8% | 2,367 | 85.2% |
How five-minute markets resolve, by price
The second half of the record, and the part anyone can reproduce. For every resolved slot we read the market price at the confirmation point — 180 seconds into the 300-second window, on the side the signal takes — and compare it with how the slot resolved. Measured on 135,904 slots. All eight bands are shown, including the two where the gap is negative.
| Price at T+180s | Slots | Avg. price | Resolved that way | Gap (pts) |
|---|---|---|---|---|
| 0.00 – 0.40 | 1,561 | 0.358 | 32.0% | −3.7 |
| 0.40 – 0.50 | 4,678 | 0.462 | 43.5% | −2.7 |
| 0.50 – 0.60 | 11,722 | 0.551 | 55.6% | +0.5 |
| 0.60 – 0.70 | 15,231 | 0.651 | 66.8% | +1.7 |
| 0.70 – 0.80 | 19,082 | 0.751 | 78.1% | +3.1 |
| 0.80 – 0.90 | 25,373 | 0.852 | 88.5% | +3.2 |
| 0.90 – 0.95 | 20,290 | 0.927 | 95.2% | +2.5 |
| 0.95 – 1.00 | 37,967 | 0.977 | 98.8% | +1.1 |
What this table is worth
It is checkable against the exchange’s own API, measured on 135,904 slots, and it requires trusting no one. The widest gap — 3.46 points between 0.80 and 0.90 — is about the same size as the roughly 3.5% it costs to take liquidity, which is why we present it as a map of where short-dated markets misprice rather than as a return.
How to check this
Every slot is one row: the asset, the five-minute window, the confirmed direction, the market price read at the confirmation point, and the outcome. The hit rate is the share of correct rows among those with a known outcome. Nothing is rounded in our favour and no month is dropped. When a number on this page changes, it changes because the measurement changed.