naijapredictions
You get the whole board here and not the good half of it, so yesterday's losing tip sits in the same column as the one that came in.

NaijaPredictions

Methodology

This page describes how a prediction here is produced, what goes into it and what it cannot do. It is deliberately specific, because a methodology page that could have been written for any prediction site in any country describes nothing.

Model variables

Five inputs go into every read. Recent form across a fixed window of fixtures rather than a window chosen after the fact. Goals scored and conceded, split between home and away, because that split behaves differently in the NPFL than in the European leagues. Head to head, used only where the two sides have met recently enough to describe the current squads. Rest days between fixtures. And the quoted price, which is an input rather than a decoration: a read that likes a team says nothing until it is set against what the market charges for that opinion.

Two things are missing and the gap is stated rather than papered over. No team news or line-up feed sits behind these pages, so a late withdrawal is not in the model. And NPFL match data is published thinner than English or Spanish data, so local fixtures are read with fewer observations behind them.

How confidence is calculated

Confidence is a rank within a single day's card, expressed in three steps. It compares the gap between our read and the quoted price across that day's fixtures, then orders them. It is not a calibrated probability and has not been checked against outcomes, and it will not be called either until an archive exists to check it against. A high label means the entry looked like the best value on the card that morning.

Limitations we accept

Prices move, and the number printed beside a pick is the one seen at one operator at one moment. Markets that settle server-side, crash and aviator rounds above all, sit outside the model entirely, since the multiplier does not exist in readable form before the round runs. Lottery draws are excluded because no live results feed sits behind this site, and a stale results page misleads more than a missing one.

Nothing here has been validated by a payout test, an app measurement or a stopwatch. Where this site quotes an operator on withdrawals it is quoting that operator's own wording, and the wordings read on 21 August 2026 disagree with each other by days.

Full published record

Every entry lands at /record/ once settled and stays there. The archive is empty on the day this page goes up, and that is the honest state of it. A methodology is worth only as much as the record that follows it, so that record is published in full, losing runs included.