📊 “Where the AI does best” series: How to use the Results filters · The AI's best sport · The best markets · The best leagues · The best combination.
Why combining filters changes the reading
So far we've looked at each dimension separately: the sport, the market and the league.
The next question is the most tempting one: what if I cross them?
Is there a league and a market where the AI doesn't miss?
The answer has a catch: every time you cross two filters, the group shrinks enormously.
A league with 70 picks, crossed with a market, can end up with 20, and with 20 picks chance rules.
How to read the numbers in 30 seconds
To pick an AI's best combination, counting hits isn't enough.
| Number | What it is | How to read it |
|---|---|---|
| Win rate | The percentage of picks that came in. | Compare it with what the odds implied (implied probability). |
| ROI | Net profit ÷ money wagered, using each pick's stake. | It's the same ROI Results shows when you filter. |
| ± (margin) | How far the ROI can move by chance alone (95% confidence). | If it's larger than the ROI, it can't be told apart from luck yet. |
| vs odds | Win-rate points above or below what the odds implied. | If it's positive, the AI beat the odds on those picks. |
| Picks | The sample: how many picks are behind each number. | More picks, more reliable. Only groups with a big enough sample enter the ranking. |
With few picks, the ROI swings wildly, which is why the ranking doesn't get carried away by a nice-looking number.
In Gambeta: reviews performance is calculated at a flat stake (always betting the same amount), so the figures don't match this series, which uses each pick's stake just like Results does.
Matching the odds is already hard: they are the toughest opponent in betting.
Gambeta manages it overall and pulls ahead of them in a few corners, which are the ones we show below with their margin of error.
🌟 Track record highlights
League and market combinations, from best to worst
Each row crosses a league with a market, for example Copa Libertadores + Home to win.
Only those with 20 or more settled picks make the list.
There are few, because when you cross two filters each group shrinks fast.
On Results, that means combining the League filter with the Market one.
| Rank | Combination | Picks | Win rate | vs odds | ROI (±95%) |
|---|---|---|---|---|---|
| 1 | Copa Libertadores + Home to win | 22 | 77.3% | +15.4 pts | +21.5% (±29.6) |
| 2 | MLB + Winner (MLB) | 111 | 58.6% | +6.4 pts | +19.7% (±19.0) |
| 3 | MLB + MLB props | 31 | 48.4% | +1.3 pts | +9.7% (±40.6) |
| 4 | World Cup 2026 + Home to win | 22 | 72.7% | +1.3 pts | +6.3% (±27.0) |
| 5 | Tennis (ATP & WTA) + Winner (tennis) | 120 | 55.8% | −0.7 pts | −2.7% (±18.2) |
| 6 | Brasileirão Série A + Home to win | 24 | 54.2% | −6.6 pts | −2.9% (±36.5) |
| 7 | La Liga + Over 2.5 goals | 23 | 47.8% | −7.4 pts | −11.6% (±38.9) |
| 8 | NFL + NFL props | 35 | 45.7% | −5.9 pts | −13.0% (±32.1) |
| 9 | Argentine Primera División + Over 1.5 goals | 25 | 60.0% | −3.4 pts | −13.9% (±33.5) |
| 10 | Argentine Primera División + Draw | 22 | 27.3% | −6.3 pts | −14.3% (±59.4) |
| 11 | Copa Sudamericana + Home to win | 23 | 47.8% | −13.3 pts | −29.4% (±36.7) |
| 12 | Liga MX + Over 2.5 goals | 23 | 39.1% | −19.5 pts | −43.3% (±34.5) |
Analysis: combinations with a positive ROI
Copa Libertadores + Home to win
🟡 Promising: needs more data
−60%0+60%
Likely ROI range: from −8.0% to +51.1%. The white dot is the current ROI; the yellow band shows how far it could move by chance alone.
In a 22-pick sample it won 77.3% and the odds implied 61.9%.
That is 15.4 points above what the odds implied.
It can't be told apart from luck yet, because the margin (±29.6) is larger than the ROI.
It was positive in both halves of the track record (+26.0% and +17.2%).
Too good to be true?
An ROI of +21.5% from only 22 picks sounds too good to be true.
That's why we flag it with an asterisk: with so few picks, a single streak changes everything (margin ±29.6).
We still show it, because we'd rather you see it with its warning than hide it.
That's the difference between a track record you can verify and one that is only advertised.
🎯 Featured pick from Copa Libertadores + Home to win
⏳ The full prediction for this pick has been taken down. Gambeta takes down its previews 20 days after the match.
The model gave it 63% and the odds of 1.51 implied 66.2%.
The odds already priced in that probability.
What was recorded for this pick
| Detail | Value |
|---|---|
| Model probability | 63% |
| Probability implied by the odds | 66.2% |
| Odds | 1.51 |
| Confidence and stake | High · $130 |
| 1X2 odds (home · draw · away) | 1.51 · 4.09 · 7.30 |
| Final score | 0-1 |
| Result | Lost · P/L −$130 |
MLB + Winner (MLB)
✅ Positive even with the margin
🌟 The best of MLB + Winner (MLB)
- Team bet on: San Diego Padres: 3 of 4 correct (75%).
The samples are small: they're for keeping an eye on, not for promising anything.
−60%0+60%
Likely ROI range: from +0.7% to +38.7%. The white dot is the current ROI; the yellow band shows how far it could move by chance alone.
In a 111-pick sample it won 58.6% and the odds implied 52.2%.
That is 6.4 points above what the odds implied.
Even after subtracting the margin of error, the ROI is still positive.
It was positive in both halves of the track record (+20.3% and +19.1%).
Too good to be true?
An ROI of +19.7% sounds too good, and doubting it is healthy.
We take it seriously for three reasons: the sample is 111 picks, it was positive in both halves of the track record and it hit more often than the odds implied.
What we can't promise yet is that it will last: its margin is ±19.0.
We publish that margin on purpose, so you can check it yourself and don't have to take our word for it.
🎯 Featured pick from MLB + Winner (MLB)
MLB + MLB props
🟡 Promising: needs more data
−60%0+60%
Likely ROI range: from −30.9% to +50.3%. The white dot is the current ROI; the yellow band shows how far it could move by chance alone.
In a 31-pick sample it won 48.4% and the odds implied 47.1%.
That is 1.3 points above what the odds implied.
It can't be told apart from luck yet, because the margin (±40.6) is larger than the ROI.
It flipped sign between the first half (−18.0%) and the second (+38.6%): it still needs more data.
Too good to be true?
An ROI of +9.7% from only 31 picks sounds too good to be true.
That's why we flag it with an asterisk: with so few picks, a single streak changes everything (margin ±40.6).
We still show it, because we'd rather you see it with its warning than hide it.
That's the difference between a track record you can verify and one that is only advertised.
🎯 Featured pick from MLB + MLB props
⏳ The full prediction for this pick has been taken down. Gambeta takes down its previews 20 days after the match.
This pick was generated from real odds from Melbet.
The odds of 2.50 implied 40.0%.
What was recorded for this pick
| Detail | Value |
|---|---|
| Odds | 2.50 |
| Confidence and stake | Medium · $30 |
| Final score | 1 bases totales (2-4) |
| Result | Lost · P/L −$30 |
World Cup 2026 + Home to win
🟡 Promising: needs more data
🌟 The best of World Cup 2026 + Home to win
- Team bet on: Argentina: 4 of 4 correct (100%).
The samples are small: they're for keeping an eye on, not for promising anything.
−60%0+60%
Likely ROI range: from −20.7% to +33.3%. The white dot is the current ROI; the yellow band shows how far it could move by chance alone.
In a 22-pick sample it won 72.7% and the odds implied 71.4%.
That is 1.3 points above what the odds implied.
It can't be told apart from luck yet, because the margin (±27.0) is larger than the ROI.
It flipped sign between the first half (−21.6%) and the second (+34.9%): it still needs more data.
🎯 Featured pick from World Cup 2026 + Home to win
⏳ The full prediction for this pick has been taken down. Gambeta takes down its previews 20 days after the match.
This pick was generated from real odds from DBbet.
The odds of 1.34 implied 74.6%.
What was recorded for this pick
| Detail | Value |
|---|---|
| Market-implied probability | 71% |
| Probability implied by the odds | 74.6% |
| Odds | 1.34 |
| Confidence and stake | Max · $170 |
| 1X2 odds (home · draw · away) | 1.34 · 4.80 · 11.00 |
| Final score | 3-2 |
| Result | Won · P/L +$58 |
Under observation: promising combinations with very few picks
Between 12 and 19 picks.
They are shown so you can follow them, not to bet on.
With such a small sample, the best one on a list is usually just luck.
| Combination | Picks | Win rate | vs odds | ROI (±95%) | |
|---|---|---|---|---|---|
| · | Premier League + Home to win | 12 | 75.0% | +13.5 pts | +31.0% (±27.6) |
| · | World Cup 2026 + Over 2.5 goals | 13 | 76.9% | +20.8 pts | +28.3% (±50.0) |
| · | Serie A + Home to win | 12 | 75.0% | +13.2 pts | +20.0% (±43.3) |
| · | Copa Libertadores + Over 1.5 goals | 13 | 76.9% | +13.5 pts | +19.7% (±39.1) |
| · | Ligue 1 + Over 2.5 goals | 16 | 75.0% | +18.7 pts | +16.0% (±53.7) |
Zones under construction: the combinations with the most room to improve
It's the flip side of the ranking: knowing where to tread carefully is also an edge.
They are combinations with a large enough sample and an ROI of −10% or worse.
| Combination | Picks | Win rate | vs odds | ROI (±95%) | |
|---|---|---|---|---|---|
| · | Liga MX + Over 2.5 goals | 23 | 39.1% | −19.5 pts | −43.3% (±34.5) |
| · | Copa Sudamericana + Home to win | 23 | 47.8% | −13.3 pts | −29.4% (±36.7) |
| · | Argentine Primera División + Draw | 22 | 27.3% | −6.3 pts | −14.3% (±59.4) |
🌟 What's already working inside
- Copa Sudamericana + Home to win: team bet on Vasco da Gama (2 of 3)
The samples are small: they're for keeping an eye on, not for promising anything.
Data as of , from the public track record (pick by pick in the open CSV). ROI = net profit ÷ money wagered, using each pick's stake: it is the same ROI the Results page shows when you filter (it doesn't include the long-term World Cup bets). The margin (±) is the 95% margin for chance: if it is larger than the ROI, the difference can't yet be told apart from luck. It updates itself every day and is not a recommendation of where to bet.
Where shouldn't Gambeta miss?
The honest answer: there is no combination anywhere that doesn't miss.
What you can do is separate where there's evidence from where there isn't.
| Traffic light | What it means | What to do |
|---|---|---|
| 🟢 Well backed | Large sample (50 picks or more), positive ROI with the same sign in both halves of the track record | It's the closest thing to a zone of trust |
| 🟡 Promising | Positive ROI, but with few picks or a sign that flips between halves | Follow it, without betting big on it |
| ⚪ No signal | ROI close to zero | Today it neither won nor lost against the odds |
| 🔴 Weak | Negative ROI with a large enough sample | Knowing where not to go in is an edge |
The best-of-the-list effect
If you take a hundred combinations and look only at the best one, it's almost certainly the one that got the luckiest.
In statistics it's called regression to the mean: over time, extremes tend to move closer to the average.
That's why combinations with very few picks are shown as “under observation” and don't enter the ranking.
A good sign is consistency: a combination that's positive in the first half of the track record and in the second is far more believable than one that was very good in only one.
Why you can trust the numbers (and check them)
A track record that only shows the good isn't useful for making decisions.
That's why we publish everything, with its margin of error, and give you the tools to check it.
| What we do | How you check it |
|---|---|
| ✅ Every pick is logged before the match | In the track record, with its odds and its date. |
| ✅ Nothing is deleted: hits and misses both stay | Results shows them all, the ❌ too. |
| ✅ The full track record can be downloaded | There's an open CSV, pick by pick. |
| ✅ Every ranking comes with its margin of error | Look at the “ROI (±95%)” column. |
| ✅ The numbers recalculate themselves every day | The date of the data is at the foot of each table. |
That transparency is the foundation of the product: what you see is what there is.
How the three filters combine on Results
- On Results, choose the Sport.
- Choose the League you're interested in.
- Choose the Market: the menu recalculates with that league's picks.
- Look at the cards at the top: Total tells you the sample and ROI the performance.
The guide to the five filters is in how to use the Results filters.
What to do with this ranking
- Use the 🔴 zones to stay out of where the AI doesn't perform yet: it's as valuable an edge as finding a 🟢 zone.
- A 🟡 zone is followed, not bet big on: wait for it to build a bigger sample.
- Even in a 🟢 zone, use a flat stake and a bankroll: no combination is safe (what a bankroll is).
- For a broader look at how to use the AI with caution, read safe sports betting.
None of this is a promise of profit: no AI hits every time.
Only play what you can afford to lose and see how to bet responsibly.
🔎 Cross the filters yourselfCombine Sport, League and Market on Results and look at each combination's sample.Open Results →Frequently asked questions
What is the AI's best combination of sport, market and league?
The one that currently tops the ranking above. The ranking goes from highest to lowest ROI, among the combinations with enough picks. A combination with many picks is worth more than one with few, so always look at the number of picks.
Where shouldn't Gambeta miss?
There is no safe combination anywhere. What you can do is tell a well-backed zone (large sample, positive and consistent ROI) from one with no evidence. No AI hits every time.
Why are there so few combinations in the ranking?
Because when you cross league and market, each group shrinks. Only combinations with 20 picks or more enter; those with fewer appear “under observation”.
What is regression to the mean and why does it matter?
It's the tendency of extreme results to move closer to the average with a bigger sample. If you pick the best combination from a long list, it almost certainly got lucky, and it's likely to do worse in the future.
How do I know whether a combination can be trusted?
Look at three things: that it has many picks, that its ROI is positive even after subtracting the margin of error, and that it was positive in both halves of the track record.
