1. Data Sources

Odds are collected continuously from Kambi-connected bookmakers — among them 1x2.se (operator ID pafpre1x2se), Unibet, and Paf — plus The Odds API as a complement for broader market coverage.

We use Pinnacle as our reference market. The reason is simple: Pinnacle has the lowest margin and the highest limits in the industry, which makes their odds widely regarded as the market's most accurate estimate of true probabilities. It's against that reference that we compare other bookmakers' odds.

Data collection runs through a multi-layer chain with automatic failover: if a source fails or responds too slowly, the system moves on to the next source in the chain. Data that, for whatever reason, is served from cache is always flagged as such in the interface, so it's never unclear how fresh a figure is.

The frontend never calls external APIs directly. Everything goes through our own backend layer, which handles fetching, validation, caching, and switching between sources. That keeps behavior consistent and traceable, regardless of which source happens to respond.

2. From Odds to Probability

The starting point is always decimal odds. The raw implied probability for an outcome is calculated as:

Implied probability = 1 / odds

If you sum the implied probability for every outcome in a market, the total always exceeds 100%. That excess is the bookmaker's margin — often called the vig or overround — and it's how the bookmaker secures a profit regardless of outcome.

Illustrative calculation example — not a historical result. We follow a match through the entire calculation chain: a 1X2 market with odds of 2.10 / 3.40 / 3.80 for a home win, draw and away win respectively.

OutcomeOddsImplied probability
Home win2,1047,6%
Draw3,4029,4%
Away win3,8026,3%
Total—103,3%

The total comes to 103.3%, i.e., 3.3 percentage points above 100%. That's the bookmaker's margin on this market.

3. De-vigging (Margin Removal)

To get to the market's actual estimate of the probabilities, the margin has to be removed. This is done proportionally, so each outcome's share of the excess is proportional to its own implied probability, and the total after removal comes to exactly 100%. The result is called estimated market probability — the market's estimate without the bookmaker's markup.

Fair probability = Implied probability / Sum of implied probabilities
Fair odds = 1 / Fair probability

For a home win in our example: 47.6 / 103.3 = 46.1%. That gives fair odds of 2.17 — the margin-adjusted odds calculated from the reference market in the example.

OutcomeImplied probabilityFair probabilityFair odds
Home win47,6%46,1%2,17
Draw29,4%28,5%3,51
Away win26,3%25,4%3,94

4. Edge Calculation

With a fair probability in hand, we can compare it against what other bookmakers actually offer. The difference — the edge — is calculated like this:

Edge% = (fair_probability × offered_odds − 1) × 100

Continuing the example: fair probability for a home win is 46.1%. Another bookmaker offers 2.30 on the same outcome. Then the edge becomes:

(0,461 × 2,30 − 1) × 100 = +6,0%

An edge of +6.0% means that the bookmaker's odds, according to our model, are higher than what the estimated market probability in the model corresponds to.

Not every positive edge gets published. Two thresholds must be met before a bet is shown to members:

  • Edge must exceed 3%. Smaller deviations fall within the normal noise of the model and the data.
  • Odds above 3.50 are not published. At such high odds, the variance in individual outcomes becomes too high for the edge figure to be meaningful in practice.
  • Edge above 8% is not published either. Such a large deviation from a sharp reference market is almost always a stale quote — either the bookmaker has already moved its line, or the reference price is outdated. Genuine edges in liquid markets sit at a few percentage points.
  • Markets without reliable reference pricing are not published: corners, shots, shots on target and team totals. Pinnacle rarely prices them, and then "edge" is measured against something that doesn't hold up.

In our example, odds of 2.30 with an edge of +6.0% clear all the thresholds and meet the selection criteria in the example.

5. Stake Level (Units)

Recommended stake is expressed in units. One unit equals SEK 100. Since 12 September 2026, the stake is flat: 1 unit per bet, regardless of how large the edge is.

Why the scaling was removed

Until then, the stake was scaled by edge: 3 units above 8%, 2 units between 5 and 8%, otherwise 1 unit. When calibrated against 406 graded bets with frozen publication values, the scaling turned out to be anti-correlated with the outcome. Same bets, same results, the only difference being the staking rule:

Scaled stake: −10.75 units ROI −1.81 %
Flat 1 unit: +8.02 units ROI +2.06 %

The entire loss therefore lay in the staking rule, not in the selection of bets. The explanation is that a scaling rule assumes a calibrated probability: if the estimate is systematically wrong, any scaling amplifies the error instead of exploiting it. We did not flip the scaling the other way — that would have been fitting to the same small sample once again. Flat staking assumes nothing at all, and is the only stance we can defend before the calibration is shown forward in time.

Being clear about the limits: the comparison above is measured on the 384 bets that had been settled when the staking ladder was removed in September 2026, not on the full history. Flat stake then produced +2,06 % with an uncertainty of ±6,2 percentage points. The sample has grown since then and the same calculation no longer produces a positive result — current figures are always shown on Results, which is recalculated for every settlement. It is not evidence that the model is profitable. It is evidence that the staking ladder cost money. The history is not rewritten — on Results you can filter by model version and compare for yourself.

6. Grading

Bets are graded automatically against real match results. There's no manual involvement in grading and no cherry-picking of which bets count.

Profit = units × (odds − 1)
Loss = −units
Void = 0

Void is used for a canceled match, an invalid market, or if the bet couldn't be graded against a reliable source for technical reasons.

Completing the example with a winning bet: 1 unit at odds of 2.30 gives 1 × (2.30 − 1) = 1.30 units in profit, i.e. SEK 130 on a stake of SEK 100.

Odds are timestamped at publication. The odds we report in statistics and results are the odds available when the bet was published — not a better price that may have appeared later.

7. CLV (Closing Line Value)

CLV measures our published odds against the market's closing odds — the last odds available before the match started. Positive CLV over time is considered the strongest indicator that a model has a possible, lasting edge according to the model, since the closing odds reflect all the information the market has had time to price in.

Being upfront: Measuring CLV requires saving closing odds at match start. BetSynq is starting to collect this going forward. Historical bets lack closing odds and can't be recalculated retroactively. CLV will therefore only be reported once there's enough data to be meaningful. We'll say it plainly, with no excuses: right now, there's no historical CLV figure to show.

8. What We DON'T Do

  • We don't hide losing bets. Every published bet is graded and reported, regardless of outcome.
  • We don't adjust odds after the fact. The odds we report are always the ones available at the time of publication.
  • We don't show returns on data where odds are missing. If there isn't enough data for an accurate calculation, we show no figure instead of a guess.
  • We don't promise profit. Positive edge is a mathematical advantage over time, not a guarantee of the outcome in any single bet.