How DeePicks turns available records into estimates.

Where the records come from

The production schedule and finished-match collection use Flashscore records. The history store also admits previously collected Sportmonks and Goal API records. Team and fixture identities are checked within their competition; conflicting or ambiguous identities can be excluded rather than guessed. Coverage is limited to the competitions and observations actually collected, not every match or every statistic in a season.

Where records can be reconciled, the history assembly prefers stored Sportmonks observations, then Goal API, with Flashscore filling missing fields. Conflicting final scores are not merged. Flashscore card fields are excluded from this field-filling step while their definitions remain unresolved. A familiar provider name does not establish that a particular fixture has complete history or that DeePicks is endorsed by that provider.

Only eligible earlier information

The scheduled pipeline limits knowledge to the earlier of generation time and two hours before the target kickoff. It checks both the historical event time and when an observation became available to the store. Shots-on-target history additionally uses dates before the target fixture's UTC calendar day. Later results and later-admitted corrections must not become earlier knowledge.

History is filtered separately for each market's required fields and target competition. It is not a universal last-five-match sample. A team can have goal records but no usable shots-on-target records. Historical counts therefore differ by market, team and snapshot.

History gates differ by market

These are minimum eligibility gates in the current scheduled pipeline, not recommended sample sizes or confidence certificates. Other required fields and checks can still make a market unavailable.

Match result, double chance and full-time goals

Require eligible full-time scores in the competition and at least one eligible appearance for each team. Such a small minimum can produce fragile estimates; availability alone is not a reason to trust a percentage.

Both teams to score

Requires at least 20 eligible competition matches with complete goal scores and at least six appearances for each team.

First-half goals

Requires at least four eligible appearances for each team with both halftime scores present.

Shots on target

Requires at least five eligible competition matches with both teams' SOT counts. The scheduled publication gate also requires represented SOT history for each target team.

Corners and cards

Use their own complete statistic pairs. The scheduled pipeline requires eligible competition history and a represented appearance for each target team. A large goals sample cannot repair a missing corner or card sample.

Different questions need different models

Full-time goals and match outcomes use a goal-count model; double chance combines the relevant match-result outcomes. BTTS addresses whether both sides score and has its own history gate. First-half goals uses halftime data rather than simply halving a full-time estimate. SOT uses observed threshold frequencies with league and team weighting. Corners and cards use separate count estimates and separate input fields.

The current cards history uses yellow-card counts. It is not automatically equivalent to a bookmaker's red-card weighting, booking points or player-card rules. Source definitions and settlement rules need separate checking. Expected goal counts in a statistical model should not be confused with a provider's shot-level xG measurement.

What missing and validated mean

Each published market is AVAILABLE or UNAVAILABLE. Unavailable reasons distinguish insufficient history, missing inputs, unavailable source data, an unavailable model and a disabled model. Missing values are not observed zeros and are not filled with another market's percentage.

The website checks the snapshot contract, required fields, permitted lines and probability ranges. AVAILABLE establishes a usable structure, not independent predictive validation. No verified accuracy, calibration, betting edge or historic return is established by this page. Lineups, injuries and other news should not be assumed incorporated unless a particular analysis establishes that fact.

Reading times and daily selections

Fixture kickoff and prediction-generation time are different timestamps. Public fixture times use Europe/Budapest, including its seasonal offset. A newly published snapshot can still contain an older individual prediction. The free-tip cards show the prediction's snapshot date; consult the full timestamp where available rather than treating a page visit as a new forecast.

Automatic free tips initially choose up to five upcoming READY fixtures, within the documented 55–78% model-probability band, with supported lines and one tip per match. Today's public selections are pinned for the Budapest date and retained after kickoff until a finished result is confirmed. They are not live in-play advice or continuously refreshed picks. A finished or unavailable selection is not replaced with a fabricated example.

Model profiles are not financial promises

The engine's Safest, Balanced and Higher Return identifiers describe different rankings within a model-only qualification band. The public labels describe these as higher-estimate, midpoint and lower-estimate profiles. They do not measure financial safety, diversification or expected return. The public free parlay contains up to three distinct fixtures, with two supported selections per fixture, when enough qualify.

Individual builder estimates do not establish a combined parlay probability. Related selections can be dependent. No current bookmaker price, stake, return or profitability calculation is supplied by those profile names.