Laptop showing data charts used to judge how reliable a hit rate is

FuturesAI Scoreboard: How Wide Is the Error Band?

Short answer: FuturesAI’s scoreboard showed 148 correct out of 321 graded predictions, an accuracy of 46.1%, at its November 11, 2026 check. A rough 95% error band on that is 40.7% to 51.6%, about 5.5 points either side. Slice the record into a 50-call subset and the band grows to about 13.8 points. Any filter built on a small slice mostly measures noise.

This is a statistics exercise on a published record. It is not investment advice, it predicts nothing, and leveraged trading can lose your entire principal.

A number on a scoreboard looks exact. It is not. It is one sample from a process with luck in it, and the honest question is how far it could drift if you replayed the same period.

Check it on FuturesAI

What We Work Out

What does the scoreboard actually report?

The FuturesAI scoreboard reported an overall accuracy of 46.1%, which is 148 correct out of 321 graded predictions, with 336 graded in total and 15 still awaiting results, at its last automatic check dated November 11, 2026.

When I checked the page, I found that it grades each prediction after 20 days and that the two counts, 321 and 336, come from different blocks of the same ledger. That mismatch is unpacked in FuturesAI Scoreboard: How to Reconcile 317 vs 332 Calls. The page does not publish win rates split by score band, so none are used here.

You can see the source yourself on the FuturesAI scoreboard.

How wide is the error band on 321 calls?

A 46.1% hit rate measured on 321 predictions has a standard error of about 2.8 percentage points, so a rough 95% band runs from about 40.7% to 51.6%. The formula is the square root of p x (1 – p) / n, then times 1.96.

The arithmetic, step by step. p = 148 / 321 = 0.461. p x (1 – p) = 0.461 x 0.539 = 0.2485. Divide by 321 to get 0.000774. The square root is 0.0278. Times 1.96 is 0.0545, or 5.45 points.

That is a normal-approximation shortcut. It is good enough for a sense of scale. It assumes calls are independent, which clustered crypto moves often are not, so the true band is probably a bit wider.

Pencil and graph paper used to work out a hit-rate error band
148 of 321 graded predictions is 46.1%, with a rough 95% band of 40.7% to 51.6% (FuturesAI scoreboard, November 11, 2026; band calculated here).

What happens when you filter to a smaller slice?

Filtering the 321 graded predictions down to a subset shrinks n, and a smaller n widens the error band: at 100 calls the rough 95% band is about 9.8 points either side, at 50 calls about 13.8, and at 30 calls about 17.8, assuming the same 46.1% rate.

Calls in the slice (n) Standard error Rough 95% band Range if rate is 46.1%
321 2.78 pts +/- 5.5 pts 40.7% to 51.6%
100 4.98 pts +/- 9.8 pts 36.3% to 55.9%
50 7.05 pts +/- 13.8 pts 32.3% to 59.9%
30 9.10 pts +/- 17.8 pts 28.3% to 63.9%

Here is the trap. Cut the ledger by ticker, by day of week, by score, by direction. Somewhere among the slices one will show 58% or 60%. With a 50-call slice, a 59.9% reading sits inside the plain-luck range.

I noticed that this is the same pattern as in FuturesAI Scoreboard: How to Judge a 1-of-5 Ticker Record, just with the numbers filled in.

Where does break-even sit inside the band?

With the scoreboard’s published payoff ratio of 1.18, the break-even accuracy before fees is 1 / (1 + 1.18) = 45.9%, which is only 0.2 points under the measured 46.1% and well inside the 40.7% to 51.6% band.

So the record cannot tell you whether the calls are above or below break-even. It can say the record is consistent with both. Add trading fees and the break-even rises, and the measured 46.1% sits closer to the wrong side of it.

Question Answer from the record
Is 46.1% above 45.9% break-even? By 0.2 points, which is far smaller than the 5.5-point band
Is 46.1% above 50%? Below, but 50% is just inside the band’s upper end at 51.6%
Can a 50-call slice prove an edge? No. Its band is about 13.8 points wide

A market-relative view, rather than a hit-rate view, is in FuturesAI Scoreboard: Do Its Calls Beat Holding the Market?.

How do you filter signals without fooling yourself?

To filter signals without fooling yourself, choose the filter before looking at results, require at least 100 graded calls in the slice, and judge the slice by whether its error band clears break-even, not by its headline rate.

  1. Write the filter rule down first. One sentence, no edits afterwards.
  2. Count the graded calls in the slice. Under 100, treat the result as a hint at best.
  3. Compute the band with the formula above.
  4. Compare the lower end of the band, not the middle, to your break-even after fees.
  5. Re-check after the next 100 calls. If the slice reverts, the filter was noise.

Boring, yes. It also saves money.

Check it on FuturesAI

Filter Rules to Keep

  • Fix the rule before you see the slice.
  • Minimum 100 graded calls per slice.
  • Use the lower edge of the band against break-even after fees.
  • Never risk money you cannot lose entirely; leveraged positions can lose the whole principal.

FAQ

How many predictions do I need before a hit rate means something?

It depends on how far from break-even you are. At a 46.1% rate, 321 calls gives a band of about 5.5 points either side, so differences smaller than that cannot be told apart from luck.

Does a 95% band mean the true rate is inside it?

Roughly. If you could repeat the same measurement many times, about 95 of 100 bands built this way would contain the true rate. It is a shortcut and assumes independent calls.

Why not just use the highest-scoring slice?

Picking the best slice after seeing the data inflates the result. The more slices you try, the more likely one looks good by chance alone.

Sources

Not investment advice. Past records do not predict future results, and leverage can lose the full principal.

๐Ÿ“Œ Hub guide: For the full earn-save-flip playbook โ€” pricing, taxes, cashback, and collectible margins โ€” see the Income & Smart Shopping Hub.

Fact-checked based on public sources as of November 13, 2026.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *