Methodology

Our football predictions are made by a goal model adjusted for each league, whose probabilities are published only after they have been validated, and a writer then adds an opinion that is labelled separately. Under our published method, every call carries a probability and a reference price, and every result, losses included, goes into a record that is never edited afterwards. The same process covers every football (soccer, for our South African readers) match we preview, for readers in Nigeria, Kenya, South Africa, Ghana and Uganda.

A useful comparison is the weather forecast, where a seven in ten chance of rain is judged over a whole season of afternoons and a single dry day settles very little. A football call deserves the same patience, and the same scrutiny. What follows sets out how a call is built, how it is tested, when we decline to make one and how the record is kept, so that you can check our working instead of taking it on trust.

What Is A Prediction On This Site?

A prediction here is a call: a selected outcome in a named market, published with an estimated probability and a reference price. The probability says how often we expect that outcome to happen, and the reference price is the odds quoted at the time, so the two can be compared directly. If the probability implies a shorter price than the one quoted, the market is paying more than our estimate says it should, and that gap is the whole reason to publish a call. A call without both numbers is only an opinion with a team name attached.

What a prediction is never, on this site, is a promise. Football is a low scoring game in which a deflection or a late penalty can overturn the better side, so even a strong call loses a meaningful share of the time. For that reason, the language of certainty has no place beside our calls, and words that present a selection as certain to win never appear, whatever the search term asks for. Strong favourites are still named as favourites, and they are given a number so that you can see exactly how strong.

18+ only. If gambling is causing concern, contact South Africa's National Responsible Gambling Programme helpline on 0800 006 008; readers in Nigeria, Kenya, Ghana and Uganda should contact a local health service for support.

How Does The Model Work, And How Is It Tested?

The model starts with interpretable baselines, which means methods simple enough that a reader can follow why a number came out as it did. The first of these is a goal model adjusted for each league: it estimates how many goals each side is likely to score, given the scoring habits of that league, and the probabilities for results and goals markets follow from those estimates. Perhaps a more elaborate machine would look more impressive. But a model that can be explained can also be questioned, and that matters more to us than appearances.

No probability is published until the model behind it has been validated. Why does that take so much care? Because a model can look stronger than it is when it is tested on data it has already seen. The data is split chronologically, with the model trained on earlier matches and tested on later ones it has never seen, and we check for leakage, where information from after kick-off slips into the inputs. Every new league is tested separately, since a model that reads one league well may misread another with different scoring patterns.

The tests themselves go beyond profit. We measure calibration, which asks whether calls given at six in ten actually land about six times in ten, alongside the Brier score and logarithmic loss, two standard measures that penalise a forecast for being far from what happened, with the latter hardest on confident errors. Returns are recorded too, but they come last in the argument. A short winning run is not proof of skill, because luck alone produces such runs regularly, and we would say the same of a short losing one.

Model Output Versus Editorial Opinion: Who Said What?

Every preview carries two voices, and they are labelled apart. The model output is the set of probabilities the tested model produced, shown as it came out, with the model version beside it. The editorial opinion is the writer's read of the match: the way the two sides match up tactically, and anything else a goal model cannot see, such as how much the fixture matters to each of them. Keeping them separate lets you weigh each for what it is, a measured estimate on one hand and a reasoned judgement on the other.

Does the writer have to agree with the model? No, and when the two differ, the preview says so in plain words. A writer may think the model is too generous to a side, for example, and will state that view with the reason, while the model's number stays in the preview untouched. It could be argued that these disagreements are the most useful part of a preview, since they show where the numbers and the football knowledge pull in different directions. The record carries the model version and the author for every call, so each can be judged over time.

When Do We Publish No Bet?

We publish no bet when a call cannot be justified, and there are three common reasons. Thin data is the most frequent, where too few reliable matches stand behind the estimate. A price that does not pay for the risk is another: the outcome may well be likely, but the odds on offer are shorter than the probability supports. And where the line-up is unknown, we wait, since a missing first choice goalkeeper or striker can move a match more than any statistic gathered before the team sheet.

A no bet is a decision, and it is recorded like any other call, in the same record and with the same fields. That matters for honesty in the numbers. A record that only showed the matches we fancied would hide how often the desk looked at a fixture and found nothing worth backing. For a reader, the practical lesson is that passing on a match costs nothing, and that an empty slip on a difficult day is a perfectly respectable result.

What Goes Into The Record, And Can It Be Changed?

The record is immutable, which means an entry cannot be altered or deleted once it is made. Each entry holds the event, the selection, the market, the model version, the author, the probability, the bookmaker and odds quoted, the time, the settlement rules, the final result, the void status and any corrections. Corrections are added as entries of their own, so the original stays visible. The settlement rules are stored with the call so that nobody, including us, can reinterpret a bet after the final whistle.

Losing calls and no bet records are shown alongside the winners, in full. Sample size and uncertainty are shown by league, because twenty calls in a competition say far less than several hundred, and a strike rate without its sample size is an advertisement more than a measurement. Previews are never rewritten after publication. Once a match ends, the settlement is added to what was already there, so the reasoning you read before kick-off is the same reasoning you can audit afterwards, whichever way the result went.

Data Tiers: Why Some Matches Get A Full Preview And Others Do Not

Not every match can be previewed to the same standard, so each competition is placed in one of three data tiers. Tier A means statistics, team news and eligible odds are all available, which supports a full preview with a probability and a reference price. Tier B means we hold fixtures, results and form, which supports limited analysis and nothing more ambitious. Tier C is a page that is held back until the data improves.

The principle underneath is that an available fixture feed never implies a reliable prediction. Knowing that two clubs meet on a Saturday afternoon is a long way from knowing enough to price the match. This is worth remembering for readers who follow leagues at home as well as in Europe. For example, the same checks apply to the Nigeria Premier Football League for readers in Nigeria, the Kenyan Premier League in Kenya, the South African Premiership in South Africa, the Ghana Premier League in Ghana and the Uganda Premier League in Uganda, as well as the CAF Champions League across all five countries. A CAF Champions League fixture would qualify for Tier A only with the full data and validation described above; a domestic league with fixtures, results and form alone would sit in Tier B; and a competition with too little reliable data would remain in Tier C. These are examples of how the criteria apply, rather than permanent tier assignments: where a domestic competition sits in a lower tier, the label describes the depth of data we hold at that moment, and says nothing about the quality of the football. A league moves up when the statistics, team news and odds behind it become deep enough to test.

How Fresh Are The Fixtures, Previews And Odds?

Fixtures are refreshed daily and again around kick-off, since dates and times do move. Kick-off times are shown in WAT, EAT, SAST and GMT together (WAT for Nigeria, EAT for Kenya and Uganda, SAST for South Africa and GMT for Ghana), so a reader in Lagos, Nairobi, Kampala, Johannesburg or Accra sees the local hour without doing the arithmetic. Previews are reviewed after material team news, meaning news that could change the call, such as a confirmed line-up. Where a review changes the view, the update is added with its time, and the original call stays in the record as it was published.

Reference odds carry a timestamp, because a price is only meaningful at the moment it was quoted. Prices move as money and team news arrive, so a number from the morning may bear little relation to what is available by the afternoon. When a reference price becomes too old to be useful, it is removed instead of being left to mislead. Before acting on any call, check the time beside the price, compare it with the odds in front of you, and treat a large difference as a reason to think again.

We review betting sites by testing each bookmaker for readers in one country and scoring it out of ten across nine areas, from registration through to the complaints route. Every score rests on a dated test, and commission never changes a score or a rank. The same process applies whether you follow football (soccer, for our South African readers) in Nigeria, Kenya, South Africa, Ghana or Uganda, although each country is tested separately and carries its own marks.

A useful comparison is the school report, where a single grade at the bottom tells a parent very little, and the marks per subject with the date of the term tell them nearly everything. A bookmaker review works in the same way. So, what follows explains each subject, how the mark is reached, where the evidence sits, and how money is kept away from the marking.

What Is A Review Here?

A review here is a tested, dated assessment of one bookmaker for readers in one country. That last part matters more than it might appear, because the same operator is a different product in Lagos than it is in Nairobi or Accra, with another currency and other ways of paying. Even the clock moves: a match that kicks off at a given hour GMT in Ghana is one hour later on WAT in Nigeria, two hours later on SAST in South Africa and three hours later on EAT in Kenya and Uganda. So, the Kenyan review of an operator and the Ugandan review of the same name are separate pieces of work, each with its own date.

Before any testing begins, eligibility is checked for each market, and the answer is recorded in one line. The review does not discuss it any further. That check decides which operators appear in which country, which is why the list for Ghana will not match the list for South Africa. In our view, this is the right order of work, since there is little point in timing a withdrawal from a site that a reader in that country cannot use.

What Is Tested, And How Is It Scored Out Of Ten?

Nine areas are tested, and the marks are out of ten. The areas are eligibility and registration, local usability, football markets and odds depth, the app, payments, withdrawals, clarity of terms, support, and the complaints route. Read in order, the list tracks a reader from opening an account to placing a bet and getting paid, and then to what happens if something needs to be put right. The table below sets out what each test looks at.

Football markets and odds depth is the area closest to our own work on prices. The test looks at the range of football markets an operator offers and at the depth of the prices behind them, because, as our guide to over 2.5 goals puts it, a price tells you how the bookmaker rates the game. Coverage checks include the Africa Cup of Nations, the CAF Champions League and domestic leagues in Nigeria, Kenya, South Africa, Ghana and Uganda. A reader who mostly bets the match result needs less depth than one who plays Asian handicaps, so the mark is best read against the markets you actually use.

Payments and withdrawals are tested separately, and they are the two marks a reader should look at first. Payments are tested with the local methods and in the local currency, which means the test uses the money a reader in that country actually holds. Withdrawals are tested with the time taken and any fees recorded. A deposit that arrives quickly says little about the journey back, which is the reason the two are never folded into a single mark.

Local usability and the app cover how the product behaves for a reader where they live. Clarity of terms asks how plainly the conditions are written, while support and the complaints route are marked apart, because asking a question and escalating a problem are different tasks. Where a mark comes in lower, the review states what was measured and what would raise it, and it leaves the judgement about the operator to the reader.

Where Is The Evidence Behind A Score?

Every score rests on a dated test, and the evidence sits behind each cell of the comparison matrix. A cell is one operator in one area, so a withdrawals mark for a bookmaker in Uganda has its own test and its own date behind it. A reader who wants to know why a number is what it is can trace the mark back to the test that produced it, which is a more useful exercise than arguing with the number itself.

It could be argued that this is more working than most readers will ever look at, and that is probably true. But, the reason for showing it is the same as in a prediction, where a probability and a reference price sit beside the call so that the reasoning can be checked. A mark out of ten is only an opinion until the test is visible, and the date tells the reader how fresh that opinion is.

Does Commission Change A Score?

No, commission never changes a score or a rank. Some links to bookmakers are commercial, meaning the publisher may be paid when a reader opens an account through them, and those links are disclosed. The marking comes from the tests described above, and the rank is set without reference to who pays what. An operator with a commercial arrangement and an operator without one are held to the same nine areas and the same marks out of ten.

Paid placements are labelled separately and are never scored. A paid placement is a position on a page that an operator has paid for, and the label is there so that it cannot be mistaken for a rank earned in testing. The simplest check on any review is whether a reader can tell which positions were bought and which were earned, and that is the standard these labels are meant to meet.

Which Promotions Are Shown?

Promotions are shown only where they are approved for the country in question. An offer displayed to a reader in Kenya has been approved for Kenya, and the same offer does not appear in Nigeria or Uganda unless it has been approved there as well. Every commercial page also carries the 18+ line and the local helpline for that country, so the reader sees both wherever a bookmaker link appears.

A welcome offer is not one of the nine areas, and its size plays no part in a score. What is marked is the clarity of terms, which covers how plainly the conditions attached to an account or an offer are written. A reader patient enough to study the terms before the headline figure is, perhaps, already ahead of the offer. Betting is for adults, and a stake should always be money that a reader can afford to lose.

How Old Is A Review, And When Is It Rechecked?

Every review carries its test date, and every review is rechecked on a schedule. Terms and eligibility are rechecked quarterly, since those are the parts of a bookmaker that can change without the product itself looking any different. The purpose of the date is that the number beside it always refers to a test a reader can place in time, in the same way that a price only means something when you know when it was taken.

So, how should a reader use all of this? Start with your own country, then read the payments and withdrawals marks before anything else, and check the test date on any mark that matters to your decision. If a number surprises you, trace it to its evidence instead of taking it on trust, which is the habit these reviews are built to encourage. The country comparison pages show the scores themselves, set side by side.

Five named house writers write the football here (soccer, to our South African readers), and every page carries the name of the one writer responsible for it. They are Tunde Akinwande in Lagos, Wanjiru Kamau in Nairobi, Thabo Mokoena in Johannesburg, Esi Owusu in Accra and Okello Mugisha in Kampala. A name on a page works, in my opinion, like the route painted on the side of a danfo: you know where it is headed before you climb in.

What follows sets out who covers which subject and how a page is researched and checked before it reaches you. The readers are in Nigeria, Kenya, South Africa, Ghana and Uganda, which is why kick-off times across the site appear in WAT, EAT, SAST and GMT together. The wider background sits on the about page.

Why Does Every Page Carry One Name?

One writer writes each page from the first line to the last, and that writer's name sits on it. Why does that matter to a reader? An opinion without an owner is hard to weigh, and a call on a title race or a derby is only useful when you can see who made it. So each name links to an author page that lists the writer's subjects and published work. Shared pages, the ones read in all five countries, are usually written from the Lagos desk, while the country pages belong to the writer who covers that country.

Who Are the Five Writers?

Tunde Akinwande is the football writer in Lagos, covering league hubs, clubs, head to heads, national teams and players. Wanjiru Kamau in Nairobi is the betting writer, so the market guides, predictions, reviews and comparisons are hers. Thabo Mokoena in Johannesburg covers the Premiership, the big three clubs and Bafana Bafana, and he keeps the South African wording on those pages, where a table is called a log.

Esi Owusu in Accra writes the Ghana Premier League, Hearts of Oak and Kotoko, the Black Stars, and the continental layer of AFCON and the CAF competitions. Okello Mugisha in Kampala covers the Uganda Premier League, KCCA and Vipers, the Cranes and the East African read on the wider game. The five are house voices with a city and a subject each. They hold opinions, and they do not claim a past they do not have.

How Is a Page Made?

Every page begins with the question readers are actually searching, and with research into what the best existing answers leave out. One writer then writes it, checking every number, date, record and name. A separate fact checker, who had no hand in the writing, checks the copy line by line. Every outside fact is sourced and dated, and the sources are informational ones such as encyclopaedias, public bodies and news sites. Where a mistake gets through, it is corrected openly, with the change stated on the page.

The live numbers are kept apart from the writing. Tables, fixtures and results are data modules that update themselves after every match, so the written layer never has to chase them. What the writers supply is the part that stays true from one season to the next: the format, the rivalries, the records with their years and the African players in the story.

What Do the Writers Never Do?

Three things are off the table for all five writers. Nobody invents first hand experience, so you will not read about a match attended or a bet placed that did not happen. Money is never promised either: a prediction is a call with a probability and a reference price, and the losing calls are published beside the winning ones. And no writer takes a knock at a club, a player or a country. Where a side is looking to improve, the copy gives the measurement and says what would change it, which is perhaps more use to a supporter than a complaint.

How Does a Voice Stay Consistent?

Each writer starts every new piece with their own published pages and the owner's feedback in front of them. That is how a view stated on one club page still holds on the next, unless the facts have changed, and how a turn of phrase does not get worn thin through repetition. A reader in Kampala or Accra who comes back for a second visit should meet the same person, holding the same standards. So, if a line ever reads unlike the writer named above it, open the author page linked beside the name and check it against the rest of that writer's work.