How to use predictapp to find value bets
predictapp calculates probabilities and fair prices; I choose which picks I make. The model has a calibration record you can measure. The way I pick is personal: match analysis, odds comparison and a lot of decisions not to bet. They are not the same thing, and this page explains both.
By Daniel, founder of predictapp. Updated 2026-11-12.
Eight questions from a subscriber
A subscriber watched my video, understood how the probabilities work and sent me eight questions that deserve a public answer. The hardest one: how do you get from a fair price to a decision that makes sense?
This page covers the app's labels, the 5% screen, parlays, match coverage, the calibration data we publish and my record of results, losses included.
My process in one minute
- 1.I start with the match result. I look for teams predictapp gives at least a 50% chance to win. That is my personal screen, not a rule for anyone else.
- 2.I don't bet against the market favorite by habit. I prefer matches where the book and the model see the same favorite but disagree on how much its win should pay.
- 3.I compare the same market. I take the model's probability, work out the fair price and check the odds my book actually offers.
- 4.I use +5% estimated expected value as a first screen. It is not a margin proven profitable on its own.
- 5.I look for reasons to pass. Squad value, shots created and conceded, finishing, the level of recent opponents, injuries and venue. A match can pass the maths and still not convince me.
- 6.I log the pick and review the result. Hit rate matters, but ROI and units won or lost matter more.
An example of step 5: in the video, AZ Alkmaar had a 71% chance and I found odds of 1.53. The estimated expected value was above 8%. But the team was converting just 0.07 goals per shot, so I left it on hold. Looking for value also means knowing when not to bet.
1. Why do so few matches show up sometimes?
Almost always, it is the calendar. Several times a season there are FIFA international windows: national teams play and many club leagues pause. The November 2026 window, for example, ran from 9 to 17 November. In those weeks the app shows fewer matches, and it is not because we dropped the rest of the schedule for lacking a good enough probability.
When the leagues come back, so does the normal volume of a club weekend. predictapp also differs from a sportsbook in three ways:
- ›Coverage: a book can offer friendlies, youth levels and many competitions our model does not cover yet. predictapp works with the competitions we have data for; we don't claim to show every match in the world.
- ›Availability: the app shows upcoming matches with a published prediction, in a 14-day window. It does not show matches that have started, been postponed or don't yet have usable data.
- ›Order and filters: sections like "For you" and the price signals help you choose what to check first. That does not mean the other matches were removed for not being value bets. You can also clear the league, country or market filters.
Coverage can grow, but I won't promise a competition or a date before it is confirmed. If you see few matches, check the international calendar first: most likely the leagues are paused.
2. What do fair odds, Value, possible mispriced markets and Pinnacle mean?
- Probability and fair odds
- Fair odds are the inverse of the estimated probability. If predictapp gives a team a 60% chance to win, its fair price is 1 ÷ 0.60 = 1.67. It is not what a book should offer: it is our statistical reference for judging the price.
- Possible mispriced markets
- The app compares the odds we saw at books with the model's fair odds. This signal means we found one or more markets where the best observed price was above our line, with extra criteria to highlight the most relevant gaps. It is an invitation to open the match and investigate, not an instruction to bet. Check when the price was seen: lines move, and the price may be gone.
- Pinnacle is also above our line
- It means Pinnacle was also offering odds above our fair price on at least one of those markets. It does not mean Pinnacle confirmed our prediction or that the outcome is more likely than the app says. It is a useful reference for comparing prices.
- Value 2.8x
- This figure on the team cards is not a 280% expected value. It is the estimated squad value: one squad is worth about 2.8 times the other. It helps me see the gap in resources, but on its own it decides neither the win probability nor the value of a bet.
3. How do I know the probabilities are well calibrated?
Calibration is not claiming we get every match right. It is checking that what we state as a probability happens about that often across enough comparable matches.
Our database keeps Trajan's historical evaluation by market and probability band, with the size of each sample. I pulled these examples on 12 November 2026:
| Market and band | Average probability | Observed frequency | Matches |
|---|---|---|---|
| Most likely 1X2 pick, 50-60% | 54.6% | 55.6% | 10,761 |
| Most likely 1X2 pick, 70-80% | 74.4% | 78.2% | 3,018 |
| Over 1.5 goals, 70-80% | 75.0% | 75.7% | 21,838 |
| Both teams to score, yes, 50-60% | 54.3% | 55.4% | 18,360 |
The "1X2 pick" rows score the most likely option among home, draw and away. These bands cover evaluations and results from 2022 to November 2026; they are not four independent tests from last week. The gaps are not zero either, and a small sample always calls for more caution.
For a more recent test that is easy to check, we published the study of 2,189 finished matches, from 28 August to 1 November 2026, across 31 competitions. Trajan's pick was right 54.1% of the time; in the 65-75% confidence band it averaged 69.4% and the outcome happened 70.2% of the time, over 248 matches. It is a different sample with its own criteria, not something to add to the table above.
What about by league? We store each match's competition, so we can evaluate by league. But the table above is organised by market and band, and it does not guarantee the same calibration in every league, market and band. A league with few matches does not support presenting a percentage as a firm conclusion.
It also helps to know how the model works: Trajan uses its own analysis, and in a small share of matches (about 6% in the published evaluation), when its confidence is low, the home, draw and away numbers are blended with the market. Goal probabilities don't get that adjustment.
A calibrated model is a good basis for judging odds. It does not prove that every price above its fair line is a profitable bet.
4. How do I calculate expected value, and what do I do with huge gaps?
With decimal odds:
Estimated expected value = (predictapp probability × offered odds) − 1
If the model says 60% and the book pays 1.80, that is 0.60 × 1.80 − 1 = +8%. It is a theoretical value, and it depends on that 60% being right for exactly that market.
In the video I went through several opportunities. These were the screens and decisions from that recording; it is not a list of current bets and says nothing about how they ended:
| Match | Probability | Observed odds | Theoretical EV | My decision in the video |
|---|---|---|---|---|
| Everton - Coventry | 59% | 1.72 | +1.48% | Pass: below my screen |
| Standard Liège - Beveren | 68% | 2.00 | +36.0% | Half a unit: doubts about poor form |
| Frankfurt - Paderborn | 55% | 2.20 | +21.0% | Half a unit |
| AZ Alkmaar - Groningen | 71% | 1.53 | +8.63% | Wait: poor shot conversion |
36%? A number like that makes me dig deeper, not bet double. It can be a real price gap, but it can also be stale odds, a pick from a different market, news the model has not caught yet or an estimate that is too optimistic.
That is why I compare several books and check whether the market still sees the same team as favorite. I don't hunt for big underdogs by habit just because the model found a gap. In the video, Atalanta was still the market favorite against Monza, even though its probabilities were very different from ours.
One check you can't skip: winning the match is not the same as winning on a handicap, double chance, or winning with over 1.5 goals. You can't take the 59% for a straight win and multiply it by the odds for a different condition. You need the probability of that exact pick. In 90-minute markets, also check that the book settles by the same rule.
5. Why a 5% screen? Should you ask for more?
That 5% is my personal pre-selection rule, not a threshold validated by a backtest that guarantees profit. It removes small gaps that the book's margin, estimation error or a price move can wipe out.
Asking for 8%, 10% or 15% does not ensure a better ROI. You get fewer opportunities and may end up with exactly the ones where the model and the market disagree most extremely. The right way to test a threshold is with real odds available before kickoff, out-of-sample periods and enough settled bets.
My rule: if it doesn't clear roughly 5%, I usually don't even start the deep analysis; if it does, I can still pass. A calculator's expected value does not decide for me.
6. Does "Build my own combo" account for how events are related?
For two picks from the same match, when we have a valid scoreline grid, yes. The system adjusts the joint probability so two clearly related events are not treated as independent. For example, "Porto wins" and "over 1.5 goals" can't be scored properly by multiplying two separate probabilities.
It has specific limits:
- ›Two events from the same match with a valid grid: an adjusted joint probability is estimated.
- ›Events from different matches: probabilities are combined assuming the matches are independent.
- ›Three or more picks from the same match, or pairs without a usable grid: the manual calculation can fall back to multiplying probabilities. That number is not a validated joint calculation when the events are related. The automatic optimizer has extra limits and can reject pairs without enough support.
- ›Markets like corners or first half: not every relationship has a validated joint model.
Also, the price predictapp calculates is a theoretical fair price, not what the book has to pay. If you use your book's bet builder, compare against its final odds for that same pick, not against multiplied single prices, because the book may settle it differently.
A parlay can look attractive because it pays more. That does not mean its real probability got better.
7. When do I go half a unit, one unit or nothing?
I don't have a validated staking formula that turns each expected value into units. It is a judgment call, and I'd rather explain it with examples than pretend there is a scientific rule behind it.
In the video, Standard Liège showed a big theoretical gap, but its poor run made me doubt: half a unit. I also went half a unit on Frankfurt. In other combos where I was more convinced I used one unit. And with AZ Alkmaar, even though it cleared 5%, I chose not to bet at that moment.
A unit is a relative measure of a pick's size, not a fixed amount of money. To compare results consistently, everyone needs to define what their unit is worth and keep that rule over time.
A model could help study a more formal staking system, but I won't present it as if it were already proven in my results.
8. What real results do I have on record?
In Coach Dex I save my picks, the odds I entered, the units and the results. This is what my account showed on 12 November 2026, with the same ROI calculation the app uses:
| Period to 12/11/2026 | Won | Lost | Hit rate | Net units | Recorded ROI |
|---|---|---|---|---|---|
| Last 30 days | 65 | 40 | 61.9% | +9.74 | +19.7% |
| Last 90 days | 150 | 122 | 55.1% | +6.48 | +4.1% |
This is a record of picks saved and settled in predictapp, manual and from the optimizer, soccer and NFL. ROI uses the odds I entered when they are valid and, if a usable price is missing, falls back to the fair price. In the 90-day window, 264 of 272 picks had a valid entered price and 8 used the fallback. It is not an external audit of tickets cashed at sportsbooks.
The 90-day window includes picks from before the model update, so it is not a pure 90-day record with the current version of Trajan. And the percentages change with every new settled pick: these numbers are for that date only.
The difference matters: the ROI of my picks measures my process, my prices and my decisions; Trajan's calibration measures the model's probabilities across many matches. Neither number replaces the other.
9. Have we shown positive CLV against Pinnacle's close?
We have not published a systematic, validated CLV (closing line value) study of my picks. Saying yes would get ahead of the data.
CLV compares the odds you got before the match with the closing odds of the same market. If you took 2.20 and the line closed at 2.00, you got a better price than the close; that is an interesting signal even if that one bet lost.
We keep historical odds in our database. That is a basis for studying CLV, but it does not mean we have already checked each of my picks against Pinnacle's last comparable price before kickoff. That needs the bet time, the market and its line pinned down, a check that a closing price exists, and an aggregate result with the sample and missing data in view.
It is a measurement I want to add: it would show whether the picks get competitive prices, regardless of the immediate result.
10. So how do I recommend using predictapp?
If you want a number that says "bet here and you'll win", that is not how I use the app. What you do get is more useful for building your own judgment: probabilities you can check, fair odds, price-gap signals, team context, parlay maths and a log of your decisions.
My advice: start with who wins the match, use the price gap to choose what to check, review the context and decide on purpose between betting, waiting and passing. I don't follow possible mispriced markets blindly, and I don't need Pinnacle to agree on every bet.
If you are a subscriber with the right access, I also share my picks before the matches, with my reasons. They are my decisions; the model gives you the tools to make yours.
Questions people ask
Does a 70% probability mean the match is a sure thing?
No. A well-calibrated 70% still misses about 3 times in 10.
Are possible mispriced markets a list of picks?
No. It is a signal that compares observed odds with our fair prices. You still need to confirm the market, the available price and the context.
Does "Pinnacle is also above our line" guarantee value?
No. It describes a Pinnacle price above our line on one market, not an independent confirmation of profit.
What does Value 2.8x mean?
That one team's squad has an estimated value about 2.8 times the other's. It does not mean a bet has a 280% expected value.
How are fair odds calculated?
Decimal fair odds are one divided by the estimated probability. At a 60% probability, the fair odds are about 1.67.
Is a +5% expected value screen proven profitable?
No. It is Daniel's personal first screen, not a guarantee of return or a threshold validated for every league and market.
Can you make money hitting only 55% of your bets?
It depends on the odds you get and the units you risk. A hit rate on its own does not prove profit.
Why do so few matches show up on predictapp sometimes?
Usually because of a FIFA international window, when many club leagues pause. predictapp covers the competitions it has data and predictions for, so the volume returns when the leagues do.
Does predictapp calculate correlation in same-match parlays?
For same-match pairs with a valid scoreline grid, yes. For combinations without that grid, and for some markets, the number assumes independence and is not a validated joint calculation.
Check the probabilities on your next match
One free reveal every day, on any match you choose.
Create free accountpredictapp is for adults 18 and over. No model can guarantee a result. Never bet more than you can afford to lose.