Take it or leave it (2025)

08.07.2025

Analysis of the results obtained

On May 15th, the Tennis algorithm entered a losing streak. Today we’re writing this article to explain what might be behind this bump on the road, and to review every tennis tournament to see whether we’re simply living through normal variance or something else is going on.

If you don’t feel like reading the whole article, here’s a brief TLDR:

The analysis (covering 1 January 2024 – 17 June 2025) shows that, while the system is still profitable in units, the profit in euros is far below expectations.

Is it just variance? A one-off blip? Is there more to it?

As we’ve discussed in the Discord server, the system is still betting with positive EV, exactly as it always has. It’s been validated by thousands of bets over 9 years, by simulations, and by the real-world results of hundreds of users.

That doesn’t stop us from having negative weeks, and—more puzzling still—longer losing spells can still come up when you’re investing in a proven +EV system.

Let’s take a close look at what’s happening.

A bit of context

In the chart below, right around pick #17,000 (roughly week 18 of 2025), the profit line in euros (blue) dips below the EV line (green).

Put simply, whenever the blue line sits above the green, luck is on our side, and when it’s underneath, it’s not.

And as you can clearly see, we’ve spent weeks during which EV kept climbing… while the real-world profit didn’t. This isn’t the first time we’ve experienced such a dip (and won’t be the last), but unlike previous downturns, this one has gone on for a while, though isn’t our worst in terms of actual profit lost.

Now, to understand why that can happen—and why it doesn’t mean the system is broken—we need to go back to basics.

Expected value

At the risk of sounding repetitive, let’s revisit the bedrock of WinnerOdds: betting with positive expected value (EV+).

EV (Expected Value) is calculated with:

EV = Bet Size * (Bet Odds * True Odds – 1)

Since recent results have been disappointing, if we assume WinnerOdds’ minimum odds aren’t precise enough, we need another way to estimate true probabilities.

By CLV (Closing Line Value) theory, the best proxy for that true probability is the one implied by the closing odds of a bet.

Using those closing odds, we first must strip out the bookmaker’s margin:

Margin = (1 / Closing Odds Player 1) + (1 / Closing Odds Player 2) − 1

True Closing Odds = 2 * Closing Odds / (2 − Margin * Closing Odds)

“True Probability” = 1 / True Closing Odds

As @12xpert explains in his excellent paper:

https://www.football-data.co.uk/The_Wisdom_of_the_Crowd_updated.pdf

For the average odds placed, I divided our users’ potential profit if all registered bets won by the total staked, plus 1:

Average Odds Placed = (Potential Profit / Total Staked) + 1

So, that factors in all kinds of odds—starting at the opening odds until they cross WinnerOdds’ minimum odds and we stop betting (provided the price is falling).

Finally, we compute each bet’s expected value as:

EV = Total Staked * (Average Odds Placed / True Odds − 1)

What happened over the past two months?

As the chart above shows, the EV+ hasn’t vanished, but the profit hit a wall.

And, true to human nature, when the red weeks pile up, doubts, headaches, and “What am I even doing here?” feelings creep in.

Much depends on how long you’ve been using the platform, how much profit you’ve banked, and your loss aversion—but losing generally stings, and nobody enjoys rough weeks.

However, looking at the data dispassionately, what we see is:

The system still keeps winning.

The system continues to identify EV+ bets with the same consistency as always.

This chart spells it out doubtlessly:

When we view results in units (Profit Units / EV Units), the picture is obvious: the theoretical system is still cruising.

The blue curve—actual units won—doesn’t just sit above the EV line (green); it stays there with a steady, even widening gap. The algorithm isn’t just working: in many stretches it’s outperforming expectations.

No sign of fatigue. No system wear-and-tear. The books haven’t “figured us out.”

That already suggests something important: the poor results some are experiencing probably don’t stem from the algorithm. The engine is still kicking.

Yet when we look at what really matters—euros—we run into reality.

Even betting with edge, there are moments when the real-world profit fails to land.

Here’s where mean regression (and then some) shows up:

This graph tracks all user’s bets over time and lets us see whether real profit (euros) walks hand-in-hand with expected value (EV units).

You can spot superb stretches—especially between bets 4,000 and 10,000—when real profit outpaced Expected Value. Everything we touched turned to gold, and the wind was on our side.

Since bet 17,000, however, we’ve experienced a pullback.

That doesn’t mean the system works in theory but not in practice. It means variance is fickle, and real life throws extra ingredients into the mix: lower average odds, smaller bankrolls, betting limits…

But what’s up with the change in slope at the end of the graph?

Look at the final stretch—after bet 18,000—and you’ll see the green line’s slope (EV) flattening.

Did you know that Bet365 reduced their liquidity in many parts of the world recently?

You can see an obvious consequence of that here: smaller stakes per bet means lower expected value per bet. The system still spits out +EV bets, but the amount you can stake on each one has lowered, damaging the overall EV curve directly.

Details like this don’t always jump out, but they surface when looking at large datasets. While we can’t definitively prove this is the sole cause, it’s a reasonable hypothesis.

Even so, if we check the staked-volume chart, we get the same feeling as with the initial profit vs. EV graph.

The line keeps climbing steadily, with no potholes or set-backs (apart from that subtle twitch at 18,000 bets). This means the system is still churning out value and opportunities. Picks arrive at the usual pace. In fact, the rising volume and EV+ lines are among the strongest signals that the algorithm is humming better than ever.

Results by tournament

So, let’s take a look at the individual tournaments to discern whether something happened big under our noses or any of them stopped being profitable.

ATP

Let’s start with the most-followed, most-liquid competition in world tennis: men’s ATP Tour.

As the chart here shows, the system has kept humming along, profiting well above the EV+ line, wobbling a bit between bets 2,500 and 3,500, then finishing with a healthy upswing.

However, if we view things in euros (Profit € / EV Units), the picture changes quite a bit.

Between bets 1,000 and 1,500 we suffered a sharp decline in real profit. From there, the system entered a prolonged flat spell, with the blue line sitting noticeably below the theoretical EV for many weeks. It’s been clawing its way back, but as of June 17th we’re still in recovery mode.

Also, zoom in and you’ll notice that EV itself (green line) has begun to slow in the last few hundred bets. That may be because we’ve staked more on ATP lately—which are especially efficient markets with lower expected value—or because our odds were a shade worse than usual, which fits these tournaments too. Either way, the result in euros has taken a toll.

The maximum drawdown chart shows that we bottomed out at roughly –€140,000 between bets 1,500 and 2,000, and we’ve been inching back toward the profit peak rather slowly.

WTA

Next up is the women’s circuit, the WTA—lower liquidity than the ATP but still very high, relatively speaking.

In units, the trajectory is excellent: the system has stayed above EV the whole way, with a clear, steady edge. True, between bets 2,500 and 4,000 we stalled for a bit, but the ending section shoots up again with conviction.

In euros, though, the story is less conventional—this time in our favor.

Really, this is a textbook case of positive variance.

By instinct we associate variance with losing money, but variance cuts both ways. Just as a cold streak can strike, so can a hot one—and that’s what we see here: a blistering start, with returns far beyond expected EV. Does that mean the system was “better” then? No. It means that, in that segment, luck (yes, it matters) rode with us.

That also explains what we see daily in Discord and why each user’s results differ.

When a system spits out 1,000 +EV bets a day, the moment each person gets on, the bookies they use, even their bankroll settings, all shape their personal graph.

Plot this on a bell curve and most users will cluster around the average, but the tails always exist: some have very bad weeks, others very good—not because they do anything different, but because variance deals the cards as it pleases.

Hence our constant focus on the average user. Amidst betting chaos, it’s the most reliable gauge of whether the system works.

The drawdown graph backs this up. The system has been down as much as €160,000 from its peak, over several extended stretches. There’s no single dramatic plunge, but a series of significant, drawn-out setbacks that have kept us from printing new highs—yet the overall numbers remain spectacular.

CHALLENGER

Here’s a curious one: the season was going fantastic, roughly mirroring WTA’s—north of €500,000 in total profit—with an almost picture-perfect upward line, well above EV… until, without warning (as life tends to go), it plummeted hard.

A quick glance at the tail end of the EV chart tells the story.

And watch what happens in euros.

While the recent dip looks like a mild setback in units—nothing major and well within normal bounds—the euro hit has been much harsher. After flirting with half-a-million euros in cumulative profit, the system strung together several losing weeks, carving a very visible hole in the close.

The blow to real profit was so steep that it erased a big chunk of earlier gains and spawned the feeling that “everything’s going wrong,” even though the algorithm still works like before—the pain of seeing euros vanish just stings more than dropping a few units.

It’s just like this meme which I love

The drawdown chart supplements the stark reality: we dropped more than €200,000 from our all-time high—a record in Challenger tournaments—in a very short time.

I’ve got several explanations for why the gap between units and euros is so extreme here, and I’ll unpack them later. Here’s a lesson we often forget or don’t really think about.

The problem here, with variance, systems, volatility… is that you never know when the good streak will arrive, or when it will leave. If we knew, we’d stop betting right before the reds or only stake on the greens. Since we can’t, the only play is to confirm everything still works—and keep betting.

ITF MEN

Now for men’s ITF, the lower-tier circuit. Here things have been far more linear and stable—no crazy spikes, but no nasty shocks either.

In units, the system has ticked along like clockwork: always above EV, no wild swings, just steady progress. That’s a very healthy profile, especially in a competition with a high, constant betting volume.

Switch to euros and, as in every case we’ve seen, the picture shifts a bit. We haven’t outperformed the EV, but we have earned steadily. There are patches of ups and downs (notably in the first half), yet from bet 3,000 onwards, the chart stabilizes and the system keeps building profit with no scary drops.

The drawdown chart underlines it further: there have been no critical moments. We scratched the €100,000 drawdown mark in a couple of stretches, but overall it’s been a smooth ride. Recoveries have been quick and stable throughout.

ITF WOMEN

The women’s ITF circuit has followed much the same path: no wild swings, a steady, predictable climb. This time, though, profit has aligned with the expected EV more tightly.

Things look good in the first chart: for much of the run we’ve been almost glued to the EV line—at times slipping beneath it. The eye-catcher is what happens at the end: a sharp rally that finally opens up some daylight in our favour.

In the euro chart the pattern is similar. While the blue curve (real profit) lags behind EV, the system keeps grinding upward. We did spend a long stint in the red, and not until after 2,000 bets did we start clawing back. From there results turn positive, though we still haven’t reached the cumulative EV.

The DD graph tells the same tale: no major shocks, but we did hit –€140,000 from the peak at one point. Even so, the final stretch shows a clear recovery, bringing losses back to very manageable levels.

And the important question: is there a clear explanation for these results?

There is an explanation. And, as usual, there isn’t one single cause but a cocktail of factors that have piled up recently.

First off, the system is still working. You can see it in the graphs: there’s edge, there’s value, and the logic holds. But that doesn’t always translate to euros at the same pace, and here’s where a few nuances kick in.

For one, lower liquidity at Bet365 Spain and other countries. Since mid-May, odds collapse faster, the maximum stake allowed is lower, and it’s harder to get money down without swaying the market. Bet365 has added stricter global limits and most bets now get routed to a trader for approval. The upshot: you stake less, so you win less, even though the expected value is unchanged.

Meanwhile, Pinnacle isn’t affected by WinnerOdds. Since we stopped monitoring Pinnacle’s odds, we’ve noticed we don’t move their prices much (especially in higher-liquidity events). Good news—value stays intact—but it also means the edge isn’t that large.

Add the stake caps at other bookmakers. Not all are as aggressive as Bet365, but the aggregate still hurts. If you can’t place our full recommended stake, euro profit naturally shrinks. There’s value, yes, but if half your bets are for 0.5 units of profit, when the system calls for 1.75 units, the euro graph won’t jump that high even if you win. This issue has always existed with betting limits, and it affects individual users unevenly.

Lastly, there’s an important thing that doesn’t usually come up. Perhaps the system performed exceptionally well initially. A blazing opening streak can warp expectations. When things regress to the mean (aka “regression to meh”), it isn’t the system getting worse—it’s the positive variance wearing off. Eventually, long-term statistical norms reassert themselves.

So yes, there’s a “why,” but it isn’t simple. It’s not that the engine broke or the bookies cracked our code. The environment keeps shifting, and that feeds straight into euro results. Yet as long as EV is alive and the average user is winning, we’re still treading the correct road.

How much good or bad luck did we have?

One of the trickiest questions when things go south is:

Am I just unlucky… or did the system bust?

To answer with data, not gut feelings, we often lean on a favourite tool of ours: Monte-Carlo simulations.

How do they work?

Assume the true probability of a player winning is roughly the inverse of the closing odds with margin removed. We can do something truly outstanding: roll the dice thousands of times—simulating every historical pick over and over, with different outcomes, yet preserving the exact same EV.

What do we see?

Each coloured line is one possible reality. They all start at the same spot and proceed with the same bets, stakes, and EV… only luck differs.

The red trace at the bottom is possible. So is the yellow rocket shooting skyward. Alternate paths in the jungle of variance.

Imagine living through the red reality (#3): treading water or losing over 10,000 bets. Or riding the yellow’s early surge (#4) only to stall later.

Every one of these realities is equally likely.

That’s hard to grasp when you’re stuck in a real downswing: you’re not special—you’re simply somewhere in the range.

So where do we stand?

Now the real deal. After 10,000 simulations, here’s the distribution of final yield after 20,000 bets.

Our real result was 3.7% yield.

And while we’re essentially in the average, the simulations tell us important details:

There was a 10% chance of surpassing 5%
As well as another 10% of ending below 1.5%

All with identical strategy. We underestimate variance because our brains can’t fathom it.

Conclusions

Often, the hardest part isn’t accepting the numbers—it’s understanding what we’re looking at. A certain optical illusion has always piqued my interest: the Müller Lyer illusion.