Nigeria audit checklist · WAT + NPFL
Audit a football prediction site in Nigeria
Audit checklist: begin with a forecast that can be reconstructed: an exact pre-match time, a permanent row, all three 1X2 probabilities and a result archive that keeps losses. Apply those checks to the combined seven-site public-evidence dataset; that dataset owns the provider list and brand-by-brand evidence comparison.
Published by Football Proof AI · Published · Updated · Non-ranking educational guide, not betting adviceLocal context · Same proof standard
Three details matter before any accuracy claim
A page can be relevant to Nigerian readers without proving that its forecasts are timely, complete or accurate. Country targeting, league coverage and model quality are separate questions.
Read publication and kick-off in WAT
West Africa Time is UTC+1. Convert both timestamps to the same zone before calculating lead time; “today”, “early” or a date without an offset is not enough to prove pre-match publication.
Name the NPFL season and fixture
For a Nigeria Premier Football League forecast, preserve the competition, season, teams, scheduled kick-off and a stable fixture identifier. A league label alone cannot join a prediction reliably to its result.
Define what “Nigeria” includes
Require the evaluation range to say whether it covers the NPFL, national-team matches, international club matches or a wider global slate shown to Nigerian visitors. Do not merge those samples under one unexplained percentage.
Timestamp test · WAT and UTC
Prove when the forecast existed
A timestamp printed beside a winning prediction can still be edited after the match. The useful question is whether a dated, externally retrievable record existed before kick-off and remains connected to the same probability payload.
- 01
Capture an exact instant
Record an ISO 8601 publication time with its WAT or UTC offset, not a relative label such as “2 hours ago”.
- 02
Normalise the kick-off
Convert kick-off to the same time zone, subtract publication time and preserve the calculated lead time.
- 03
Bind the payload
Keep the teams, league, probabilities, model version and record hash together so a later number cannot inherit an earlier timestamp.
- 04
Distinguish self-proof
A site's own timestamp and hash help detect changes, but neither independently proves that the record was public at that time.
Use the prediction receipt verifier to reproduce a declared payload hash. Its pass result verifies only the named local checks; it does not create an independent historical timestamp.
Complete record · Losses stay visible
Audit the denominator before the percentage
A defensible result page explains which forecasts were eligible, how many settled and why any row was excluded. Start from the archive and rebuild the denominator instead of trusting a badge.
- Eligible forecasts
- The declared league, market, model version and date range before results are inspected
- Settled rows
- Wins and losses under the same rule, with void, postponed and pending rows labelled separately
- Missing rows
- Any gap between public daily forecasts and the result archive, explained rather than silently ignored
- Declared denominator
- Correct count divided by all eligible settled predictions—not selected winners or only high-confidence calls
- Probability score
- Brier score or log loss calculated from the complete probability split, with the exact convention named
- Sample uncertainty
- An interval and sample-size warning so a short streak is not presented as stable model skill
Paste a stated hit rate and counts into the PROOF-5 claim checker to recalculate the percentage and record which evidence gates remain incomplete.
Probability test · Not a promise
Require home, draw and away together
A selected outcome such as “home win” hides how close the alternatives were. A complete 1X2 forecast exposes all three values, their total and the uncertainty behind the top call.
- 01
Sum the split
Home, draw and away should form one declared probability distribution and total approximately 100% after stated rounding.
- 02
Keep the original decimals
Store enough precision to score the forecast; a rounded badge should not replace the underlying values.
- 03
Check calibration
Across a sufficient sample, events labelled near 60% should occur near 60%—not necessarily win every next match.
- 04
Name the market
A 1X2 hit rate cannot be compared directly with BTTS, over 2.5 or correct-score accuracy.
Read the football probability guide before comparing percentages from different markets or models.
Method and incentives
Separate how a forecast is made from how it is sold
A method page should identify the target, data cutoff, model or editorial process and validation design. A disclosure page should separately explain subscriptions, advertising, affiliate links or other commercial relationships when they exist.
Ask what information was available
Features such as form, tables, line-ups or odds need an as-of cutoff. Later information cannot be used to justify an earlier NPFL forecast.
Ask how time was respected
Training and test periods should move forward through time. Randomly mixing future matches into training can overstate apparent performance.
Ask who benefits from the click
A commercial relationship does not prove a forecast wrong, but hiding it prevents readers from evaluating the incentive behind a recommendation.
Apply the checklist · Keep one evidence source
Use the seven-site dataset as the provider evidence source
The shared Nigeria-and-Kenya index owns the named provider list, comparison matrix and cited source dossiers. This country page supplies the WAT and NPFL audit procedure only, so the underlying observations have one canonical source instead of seven repeated profile lists.
Interpretation boundary
Evidence availability is not a profitability verdict
A located archive or method page answers one narrow retrieval question. It does not establish that predictions are accurate, calibrated, suitable for a user or capable of producing a profit. A null means only that qualifying public evidence was not located at the recorded review time—not that it cannot exist.