FuturesAI Scoreboard: How to Judge a 1-of-5 Ticker Record
Anyone who reads a prediction scoreboard ends up doing the same thing: scanning for the worst names and the best names. The per-ticker table on the FuturesAI prediction scoreboard invites exactly that. It lists XRP at 3 of 7, then XLM, BCH and DOGE at 1 of 5 each.
This is a filtering question, not a prediction. It is not investment advice, and nothing here says any ticker will rise or fall.
Contents
- What does the ticker table actually show?
- How likely is a 1-of-5 record by chance?
- How many calls does a ticker need before you can read it?
- Why does the independence assumption break here?
- What filter is worth using instead?
What does the ticker table actually show?
The FuturesAI scoreboard’s per-ticker table lists only tickers with five or more scored predictions, and on November 11, 2026 it showed four rows: XRP at 42.9% (3 of 7), plus XLM, BCH and DOGE at 20.0% (1 of 5 each). The overall record on the same page was 45.7% accuracy, 145 right out of 317 predictions. By market, crypto stood at 52.7% (48 of 91), US stocks at 45.9% (51 of 111) and Korean stocks at 40.8% (53 of 130).
When I checked the page, the scoring rule was stated plainly: the frozen price is the close when the call was made, and the resolved price is the close on scoring day. That makes each call a simple hit or miss, which is what lets us use basic probability here.

How likely is a 1-of-5 record by chance?
If a crypto ticker’s calls hit at the scoreboard’s crypto rate of 52.7%, the chance of seeing 1 hit or fewer in 5 scored calls is about 15.5%, or roughly 1 in 6.5. Here is the same calculation for each listed ticker, using the crypto hit rate of 48 of 91 as the baseline.
| Ticker | Record | Chance of this or worse at 52.7% |
|---|---|---|
| XLM | 1 of 5 (20.0%) | 15.5% |
| BCH | 1 of 5 (20.0%) | 15.5% |
| DOGE | 1 of 5 (20.0%) | 15.5% |
| XRP | 3 of 7 (42.9%) | 44.0% |
XRP’s 42.9% looks bad next to 52.7% but is a near coin flip: 44% of the time a ticker with a true 52.7% rate would do that badly or worse over 7 calls. That is not evidence of anything.
The method is a plain binomial count. For 1 or fewer hits in 5 calls: 0.473 to the 5th power (0.024) plus 5 times 0.527 times 0.473 to the 4th power (0.132). Add them and you get 0.155.
How many calls does a ticker need before you can read it?
A hit rate measured on 5 calls carries a rough 95% margin of about ยฑ44 percentage points, so a ticker needs dozens of scored calls before its percentage means much. The margin shrinks with the square root of the sample. These figures use the overall 45.7% rate and a normal approximation, which is crude at tiny samples but shows the scale.
| Scored calls | Rough 95% margin (points) | Where it appears |
|---|---|---|
| 5 | ยฑ43.7 | XLM, BCH, DOGE |
| 7 | ยฑ36.9 | XRP |
| 30 | ยฑ17.8 | no ticker row yet |
| 91 | ยฑ10.2 | whole crypto market |
| 317 | ยฑ5.5 | all predictions |
Read that bottom row carefully. Even the full 317-call record carries about ยฑ5.5 points of noise, so 45.7% sits statistically close to a coin flip plus a little. A ticker row with 5 calls tells you almost nothing.
Why does the independence assumption break here?
The 15.5% figure assumes each call is an independent coin flip, which does not hold when four crypto tickers are scored over overlapping windows. If all three 1-of-5 tickers were independent, the chance of all three landing at 1-of-5 or worse would be 15.5% cubed, about 0.37%. That looks damning until you notice the assumption is wrong.
Crypto assets tend to move together. One broad sell-off can sink every UP call in the same window at once. So the three records are really one bad stretch counted three times. And the table only lists tickers with at least five scored calls, which is itself a selection. I could not tell from the page whether rows are sorted by accuracy, so I would not treat the list as a random sample either.
Here is where people get tripped up. They see three matching 20% rows, multiply the probabilities, get a tiny number and conclude “this AI cannot do altcoins.” The multiplication is not valid.
What filter is worth using instead?
A better filter for FuturesAI signals is one that needs a large sample, such as the market-level split, rather than ticker rows with 5 to 7 calls. Use these rules of thumb:
- Ignore any hit rate with fewer than 30 scored calls. Treat it as unknown.
- Compare the market-level record, and the gap against buying the market, before comparing tickers. The page shows crypto’s predicted trades at +2.45% against +5.87% for the market, a gap of โ3.42 points.
- Re-check after more calls resolve. The page shows 15 pending predictions, so the numbers will move.
- Remember the page states no fees were deducted from its return figures.
For the sample-size logic behind rule 1, see FuturesAI Hit Rate: How Long a Losing Streak Gets. For a market-level luck test, see FuturesAI Scoreboard: Is the Crypto 55.8% Just Luck?
Tips
- Write the denominator next to every percentage. “20%” and “1 of 5” feel different.
- Screenshot the table with its date. The record changes as predictions resolve.
- Test any rule you build from this table on calls made after the date you built it.
Warnings
- This is not investment advice. Futures and leveraged products can lose the full principal.
- FuturesAI makes no profit promise here, and neither does this post. The scoreboard’s average return per call was +0.03% on 317 calls, 2.28 points below the index over the same windows.
- A low sample can flatter as easily as it can hurt. A 5-of-5 ticker deserves the same suspicion.
FAQ
Is a ticker with a 20% hit rate worth avoiding?
Not on 5 calls. A ticker whose true rate matches crypto’s 52.7% would still post 1 of 5 or worse about 15.5% of the time, so the record alone cannot justify avoiding it.
How many scored calls make a ticker record readable?
About 30 gives a rough margin near ยฑ18 points, and 100 gives about ยฑ10. Below 30, treat the percentage as unknown rather than low or high.
Does the scoreboard include trading fees?
No. The page states that fees are not deducted from its return figures, so any real result would be lower.
