How to Read a Data Quality Score — and Why It Changes the Bar

A probability with no reliability signal attached is a trap. 58% derived from twelve seasons of history, five price sources, xG and confirmed lineups is a fundamentally different object from 58% derived from results alone in a league nobody has modelled.

Both print as "58%". The data quality score is what separates them.

The composition

The score is built from weighted sources with fixed weights:

Source	Weight	Saturation point
Match history	35	around 40 matches per team
Odds depth	15	around five price sources
xG availability	15	present or absent
Lineups	15	confirmed or not
Player data	10	coverage-dependent
Injuries	5	reported or not
News	5	available or not

Two design decisions worth noticing.

History dominates. At 35 of 100 points, match history is by far the largest single contributor — because everything structural in the model, from strength ratings to attack and defence ratios, is built from it. And it saturates: past roughly 40 matches per team, more history adds nothing to the score, because the estimate is already stable.

Price depth is a data source, not a model input. Odds depth contributes 15 points to data quality because more sources make the market estimate more stable and therefore the comparison more meaningful. This is separate from — and does not contradict — the rule that market prices are never a feature in the model itself.

The benchmark: results and two books scores about 41

This is the number worth memorising, because it describes a very large share of world football.

A fixture with a reasonable results history, two price sources, no xG, no lineup feed, no player data, no injury reporting and no news coverage scores approximately 41 — labelled "limited".

That is not a broken fixture. It is a fixture where the structural model works and everything that refines it is missing.

Below 40, the fixture is flagged data-limited, and that flag is displayed rather than buried.

What a low score actually changes

It does not change the probability. It changes the bar.

The uncertainty budget — the model-versus-market gap required before a fixture is called actionable — grows with each measurable source of uncertainty, and data quality is one of them.

Scenario	Required gap	Outcome for a raw +3pp edge
High data quality, lineups confirmed, stable market	low	Actionable
Lineups unconfirmed, two hours to kickoff	higher	PASS
Lineups unconfirmed and the market is moving	higher still	PASS
Thin data, two price sources, no xG	highest	PASS

Same gap. Four different decisions. The bar is a property of the surroundings, not of the gap.

The underlying rule: unknown information counts as uncertainty, never as certainty. Many systems treat missing data as neutral. It is not neutral, it is missing.

Data quality is not confidence, and neither is probability

Three numbers, routinely conflated:

Probability — what the model thinks will happen on the pitch.

Confidence — how stable that estimate is, computed as:

confidence = 100 × (0.35·agreement + 0.30·calibration reliability
                  + 0.20·data quality + 0.15·sample)

Data quality — how complete the underlying information is, and one of the four inputs to confidence at 20% weight.

You can have a high probability with low confidence — most obviously when lineups are unknown. You can have moderate probability with high confidence, which is often the more useful situation of the two.

Where the score gates output entirely

In two places the score does not just adjust a bar, it stops a number being published.

Goals markets require history. Over/under and both-teams-to-score are derived from the score matrix, which is built from estimated scoring rates. Those markets are published only when both teams have at least 20 matches of history. Below that, the rates carry an error bar too wide for the matrix to mean anything.

The DATA_INSUFFICIENT decision state. One of thirteen decision states exists precisely for fixtures where quality falls below the threshold for standing an estimate on. It is an answer, not a failure — and it is different in kind from "no edge", which means the analysis ran and found agreement.

Reading it in practice

Three habits that make the score useful rather than decorative.

1. Read it before the probability, not after. Once you have absorbed a number, the reliability signal arrives too late to change how you feel about it.

2. Ask which component is missing. A fixture scoring 55 because lineups are not yet confirmed will improve in an hour. A fixture scoring 41 because the league has two price sources will not improve at all. These call for different behaviour: wait, versus widen coverage or step away.

3. Treat low-quality fixtures as coverage candidates, not avoidance candidates. If you are filling a form where every match must be marked, a fixture with a flat distribution and a low quality score is exactly where paying for a second selection beats trying harder to guess.

What the score does not capture

Honest limits:

• It does not measure whether the available data is correct, only whether it is present. A feed that reports accurately and one that reports badly score the same.
• It does not capture local knowledge. A person who watches every match in a small league may know things no feed carries. In low-coverage leagues, the relative value of that knowledge is highest — which is exactly where our score is lowest.
• It is not a forecast of accuracy. A high score means the estimate rests on more; it does not promise the estimate is right.

Why publish it at all

Because the alternative is worse. A product that emits a confident-looking probability for every fixture in the world — including leagues with three months of history and one price source — is manufacturing confidence rather than measuring it.

Publishing the score, flagging fixtures below the threshold, withholding goals markets without sufficient history, and carrying an explicit DATA_INSUFFICIENT state are all versions of the same commitment: where there is no basis, there is no number.

That commitment costs coverage. It is the reason a typical round produces far fewer actionable fixtures than a content schedule would like, and it is the point.

Three habits worth adopting

1. Read the score before the probability. Once a number has landed, a reliability signal arriving afterwards is too late to change how it feels.
2. Ask which component is missing. A fixture at 55 because lineups are pending will improve within the hour. A fixture at 41 because the league has two price sources will not improve at all.
3. Treat low-quality fixtures as coverage candidates. Where every fixture must be marked, a flat distribution with a weak score is exactly where paying for a second selection beats guessing harder.

---

18+. WinPIQ is an analysis tool, not advice and not a promise. Betting can be addictive and money can be lost. Only stake what you can afford to lose, and if betting stops being entertainment, seek help. WinPIQ is not affiliated with Winner or the Israeli Council for the Regulation of Sports Betting.