Tracked, not claimed: the case for publishing every result

There are two kinds of track record in this industry. One is tracked. The other is claimed. They can look identical on a website, and they are not the same thing at all.

Two kinds of record

The most important distinction in this whole field is also the easiest to miss, because the two things it separates can be made to look the same.

A claimed record is a set of results presented as evidence of skill. A tracked record is also a set of results presented as evidence of skill. On a web page, with the same charts and the same green and red, you often cannot tell them apart at a glance. But underneath, they are opposites. One is assembled to persuade. The other is recorded to be tested. The rest of this article is about how to tell which is which, and why Axia builds only the second kind.

The easiest record to produce

The easiest impressive record in the world is a curated one, and it is worth walking through exactly how easily it is made, because the ease is the problem.

Run analysis on a few hundred matches. Wait for the results to come in. Then publish the ones that came good, frame the winners as the story, and let the rest quietly disappear. Nothing in that process is technically a lie. Every selection shown really did win. The screenshots are real. And the record is still worthless as evidence, because a record assembled after the results are known tells you only one thing: that the person assembling it can read a results page.

This effect has a name in statistics: survivorship bias. It is the error of judging a process by the examples that survived a filter, while the examples that did not survive are invisible and uncounted. If you only ever see the winning selections, you are not seeing a track record. You are seeing the output of a filter, and the filter was applied after the fact.

This is the quiet problem with a great deal of prediction content. The confident write-up, the run of green ticks, the screenshot of a good week, the bold percentage on a homepage. Very little of it is anchored to anything that was committed before the matches were played. It is a claim wearing the clothes of a record, and because it costs nothing to produce, the market is full of it.

What tracked actually means

A tracked record has one property a claimed record cannot fake, and everything else follows from it: the timestamp.

Every Axia recommendation is logged before kick-off. The selection, the price available at the time, the market, the stake, all of it is fixed and recorded while the outcome is still genuinely unknown. When the match is played, the result is settled against what actually happened, and the line stays in the record whichever way it went. Nothing is added afterwards. Nothing is removed afterwards. Nothing is reworded once the result is in.

That sequence, commit first, settle second, is the entire difference, and it is not a small operational detail. It is the thing that converts a set of results from decoration into evidence. A claim is written after the event and can therefore prove nothing, because anything can be written after the event. A tracked record is written before the event and can therefore prove everything it contains: the wins, the losses, the flat months, the strike rate, the drawdown, the average price, the lot. The discipline of logging before kick-off is unglamorous and it is the foundation the entire published record stands on.

It also constrains us in a way that is worth being explicit about. Because the record is committed in advance, we cannot improve it after the fact. A bad month cannot be revised. A poor run cannot be smoothed. The 2025-26 season returned 15.3 percent and we would not have been able to report a different number even if we had wanted to, because the lines were already written. That constraint is not a limitation of the system. It is the system.

Why we publish the losing months

The hardest part of a tracked record to publish is the part that lost, and the test of whether a record is genuinely tracked is whether the losing part is there at all.

The 2025-26 season had an October that fell 15.9 percent, the only losing month in ten. It would have been easy to let that month be quiet, to lead with the season figure and the strong spring and let October blur into the background. We publish it at exactly the same size as the strongest months, for two reasons.

The first is that a record is not honest in pieces. If October can be hidden, then anything can be hidden, and the moment a reader knows you would hide a bad month, every good month you show them is worth less. Honesty in a record is not divisible. It is a property of the whole thing or it is absent from all of it. A record is trustworthy entire, or it is not trustworthy.

The second reason is that the losing month is doing real analytical work. Genuine edges in efficient markets are small, small edges produce uneven results, and uneven results contain losing stretches. A record with no bad months has not escaped that logic; it has hidden the evidence of it. October, shown plainly, is part of what makes the rest of the season believable. We set out the full reasoning in what a 16 percent October taught us about a real edge, and the complete month-by-month picture sits in the 2025-26 season in numbers.

The problem with a record that has no bad months

It is worth stating the consequence of all this as bluntly as it deserves, because it inverts most people's instinct.

A track record with no losing months is not reassuring. It is a warning sign.

Most people read an unbroken run of green as the strongest possible evidence. It is closer to the opposite. A genuine edge in a competitive market is small, and a small edge cannot produce an unbroken run of winning months across a long enough period. So a perfectly smooth record has only a few possible explanations, and none of them is "an unusually good model." Either it has not run for very long. Or it is being shown to you with the bad parts removed. Or it is not being measured honestly in the first place. The smoothness itself is the tell.

A real record looks like the 2025-26 season: mostly up, with a clear month down, a near-flat month, two strong months, and a drawdown in the middle. The unevenness is not a flaw in the evidence. The unevenness is the evidence.

How to tell a tracked record from a claimed one

If the two can look alike on a page, a reader, and certainly an investor doing diligence, needs a way to tell them apart. Here is what genuinely separates them, and what is worth checking.

The first test is the timestamp. Was each selection committed before the event, in a form that could not be edited afterwards? A record that cannot answer that question is a claim, regardless of how it looks.

The second test is completeness. Are the losing selections and the losing months present, at the same prominence as the winning ones? A record that shows only wins, or shows losses only in small print, has been filtered.

The third test is consistency of measurement. Is every selection counted the same way, to the same stake, with no results quietly weighted heavier than others? Axia measures every recommendation to a flat, level stake for exactly this reason: it makes the record impossible to inflate selection by selection.

The fourth test is reconciliation. Do the headline figures actually add up from the underlying lines? A genuine tracked record can be reconciled from its own data. Every number in our season review derives from the recommendation log and can be traced back to it.

A record that passes those four tests is evidence. A record that cannot is decoration. The tests are not difficult to apply, and the willingness to invite them is itself a signal.

Why claimed records persist

If a tracked record is so clearly the better evidence, a fair question is why the industry is still full of claimed ones. The answer is not that nobody has noticed the difference. It is that three things keep the claimed record alive, and naming them is part of understanding the opportunity.

The first is cost. A claimed record is almost free to produce. It requires no discipline of logging before kick-off, no settlement process, no commitment to publish the bad months. You wait, you select, you present. A tracked record requires the opposite of all of that, sustained for years. The cheaper option will always be more abundant, simply because it is cheaper.

The second is speed of reward. A claimed record can look impressive on its first day, because it can be assembled from results that have already happened. A tracked record cannot look impressive on its first day, because on its first day it has one line in it. It only becomes persuasive slowly, as the sample grows. A business that needs to look credible immediately is tempted, every time, toward the version that can fake a head start.

The third, and the most important, is that in an immature market customers genuinely cannot tell the two apart. When an audience has not yet been through enough cycles, a confident claimed record and a genuine tracked record look identical, and the cheaper one wins. This is why claimed records are most abundant exactly where markets are youngest, and why they thin out as a market matures and its customers learn what to check.

That last point is the one that matters for Axia. The persistence of the claimed record is a feature of an early market, not a permanent state. As the market matures and customers learn to ask for the timestamp, the completeness and the reconciliation, the cheap option stops working. A company that built the expensive, honest version from the start is not at a disadvantage during that shift. It is the thing the shift moves toward.

Trust is the asset

It would be possible to read all of this as a statement of principle, and it is one. It is also a straightforward commercial position, and the two are not in tension.

Axia earns through subscriptions to analysis. People do not renew a subscription to analysis they have stopped believing. So the single most valuable thing the company owns is not any individual model output, and it is not even the model itself. It is the credibility of the record the outputs are written into. A tracked record, including its worst month, is what makes that credibility real rather than asserted, and credibility, once it is genuine, is extremely hard for a competitor to copy, because it can only be accumulated in real time.

The market is moving the same way. As sports betting matures and regulation raises the standard for how products are marketed and how claims are substantiated, a curated record stops being clever and starts being a liability. An open one stops being a risk and starts being the reason a serious customer, partner or investor chooses you. Transparency is shifting from a cost to a moat, and a company that adopted it early is well placed for that shift.

What this asks of us

A tracked record is a commitment, and it is worth being honest that it is not always a comfortable one.

It means every poor month is published. It means we cannot present a difficult stretch as anything other than what it was. It means the record constrains the marketing, rather than the marketing shaping the record. A claimed record is easier in every short-term sense, which is precisely why so much of the industry runs on one.

We track everything, and we claim nothing we have not tracked, because the harder record is the only one worth reading, and because a company preparing to be trusted with other people's confidence should be measured by the standard it volunteers for when no one is forcing it. The tracked record is that standard, and we intend to keep being held to it.