FuturesAI Signal Stacking: Does More Reasons Mean Better?
Every call on FuturesAI’s public scoreboard carries a small gauge badge with a number from 2 to 5. Click it and a list opens: RSI low, volume climax at lows, outperforming index, and similar tags. The number is simply how many of those reasons were attached to the call.
It is tempting to read that badge as confidence. Five reasons must beat three, right? I wanted to see whether the ledger agrees, so I tallied it and ran the numbers.
Jump to a Section
- What does the reason count on a call actually mean?
- Does a call with more reasons hit more often?
- How big a gap can luck alone produce?
- How many calls would it take to prove a difference?
- A Reusable Check Before You Trust a Count
What does the reason count on a call actually mean?
On FuturesAI’s scoreboard, the number on a prediction’s gauge badge equals the count of signal tags listed under “Signal basis” for that call, from 2 to 5 in the ledger read on November 8, 2026. I found that every badge in a saved copy of the page matched the number of tags in its pop-up list (90 of 90).
That is a count of reasons. It is not a probability, and the page does not say it is calibrated to one. A hit, per the scoreboard, means the price direction matched the call between the frozen price and the scoring-day close. Fees are not subtracted.
Worth keeping in mind: tags overlap. “RSI low” and “volume climax at lows” often fire together because both describe a washed-out chart. Five overlapping tags can carry less new information than three independent ones.
Does a call with more reasons hit more often?
On the FuturesAI scoreboard ledger read November 8, 2026, calls with four or five reasons hit 13 of 26 times (50.0%), while calls with two or three reasons hit 27 of 62 times (43.5%). The ledger shows the 100 most recent scored records, and 88 of them display a reason count. Their prediction dates run from late August to early November 2026.
| Reasons on the call | Hits / calls | Hit rate | 95% range (my calculation) |
|---|---|---|---|
| 2 | 2 / 6 | 33.3% | 9.7% to 70.0% |
| 3 | 25 / 56 | 44.6% | 32.4% to 57.6% |
| 4 | 5 / 12 | 41.7% | 19.3% to 68.0% |
| 5 | 8 / 14 | 57.1% | 32.6% to 78.6% |
| 2 or 3 combined | 27 / 62 | 43.5% | 31.9% to 55.9% |
| 4 or 5 combined | 13 / 26 | 50.0% | 32.1% to 67.9% |
| All 88 numbered calls | 40 / 88 | 45.5% | 35.5% to 55.8% |
Data: FuturesAI scoreboard ledger, read November 8, 2026. Ranges are Wilson 95% intervals computed by me, not FuturesAI figures.
Look at the shape. Four reasons did worse than three. Five did best, on 14 calls. A real “more reasons is better” effect would climb step by step, and this doesn’t.
When I compared the two combined rows, I noticed that both ranges swallow each other almost whole. The 2-or-3 range runs 31.9% to 55.9%. The 4-or-5 range runs 32.1% to 67.9%. Overlap that heavy is the signature of a difference nobody can tell from noise yet.

How big a gap can luck alone produce?
Two groups of 26 and 62 calls drawn from the same 45.5% hit rate would differ by 6.5 percentage points or more in a bit over half of random draws, because the standard error of the gap is about 11.6 points. Here is the calculation, step by step.
- Pooled hit rate: 40 / 88 = 0.4545.
- Standard error of the gap: the square root of 0.4545 x 0.5455 x (1/26 + 1/62), which is about 0.1163.
- Observed gap: 0.500 – 0.435 = 0.0645.
- z = 0.0645 / 0.1163 = 0.55.
A z of 0.55 is nowhere near the 1.96 that conventional testing asks for. Put plainly: if reason count did nothing at all, you would still see a gap this big or bigger a majority of the time.
Honestly, that is the whole finding. It is not “the badge is useless”. It is “this sample cannot tell”.
How many calls would it take to prove a difference?
To detect a 6.5-point hit-rate gap between two groups near 50% with 80% power at the 5% level, you need roughly 930 graded calls in each group, about 1,860 in total. The formula is n = 2 x (1.96 + 0.84)^2 x 0.25 / 0.065^2, and it is my rule-of-thumb calculation, not a FuturesAI claim.
| Gap you want to detect | Calls needed per group |
|---|---|
| 13 points | about 230 |
| 10 points | about 390 |
| 6.5 points (this ledger) | about 930 |
| 3 points | about 2,180 |
The scoreboard says 325 predictions have a final result in total, across all markets. Even the entire record is far short of 1,860 calls with a clean split by reason count. So the honest status today is “unproven either way”.
The break-even bar matters too. With the page’s payoff ratio of 1.15, a signal needs a hit rate of about 46.5% just to cover itself before fees (1 / 2.15). The 4-or-5 group clears that line at 50.0%, but its 95% range starts at 32.1%, far below it.
A Reusable Check Before You Trust a Count
Run any feed that stacks reasons through this routine, with the numbers written down before you look at the result.
- Split the calls by reason count into two buckets with at least 50 graded calls each. Fewer than that and the ranges are too wide to read.
- Work out each bucket’s hit rate and a 95% range.
- Compute the gap and its standard error, as in the four steps above.
- Compare each bucket’s range against break-even, 1 / (1 + payoff ratio).
- Only act on a count if the gap clears z = 1.96 and the better bucket’s low end clears break-even.
For the full three-gate version of this filter, I walked through it in FuturesAI Signals: A 3-Gate Filter Before You Trust One. The sample-size side is covered in FuturesAI Signals: Filter Them With Real Hit Rates.
Tips
- Screenshot the ledger on the day you read it. The page shows only the latest 100 records, so old rows scroll away.
- Count ties honestly. A call that moves 0.2% your way is a hit in the ledger and almost nothing in your account.
- Rerun the table monthly. If the gap stays near 6 points as the sample grows past 900 per group, it starts to mean something.
Warnings
- Do not size up a trade because a call shows five reasons. The data here does not support that.
- The scoreboard’s returns are measured without fees. Trading costs shave the edge further.
- Leverage magnifies a wrong call by the same multiple as a right one. Leveraged futures can lose the full principal.
- This is arithmetic on published numbers, not advice to buy or sell anything. Past hit rates do not guarantee future results, a point FuturesAI makes on its own page.
FAQ
Does a higher reason count make a FuturesAI call more likely to hit?
Not that the public data can show yet. On the scoreboard ledger read November 8, 2026, calls with four or five reasons hit 13 of 26 (50.0%) and calls with two or three hit 27 of 62 (43.5%), a gap well inside what chance produces at those sample sizes.
What hit rate does a signal need to break even?
The break-even hit rate is 1 divided by (1 plus the payoff ratio). At the scoreboard’s published payoff ratio of 1.15, that is about 46.5%, before any trading fees or slippage.
Can I trust a 50% hit rate from 26 calls?
No. A 95% Wilson range around 13 hits in 26 calls runs from about 32% to 68%, so the sample is consistent with a coin flip and with a clearly profitable signal alike.
Sources
- FuturesAI AI Prediction Scoreboard (ledger and totals read November 8, 2026; last auto-check shown as 2026.11.04)
- FuturesAI Payback page (platform description and “not investment advice” notice)
