TRANSPARENCY · PREMIUM · RISK MANAGEMENT
Premium V4: how we adjusted protection and stake after auditing 69 picks
The new strategy doesn't try to get more matches right after knowing the result. It changes how much is exposed in each scenario and when it's worth pairing the 2-0 with a moneyline entry.
A strategy can end up profitable and still be over-exposing. That's what we found when reviewing the tiered execution of the BO3 Premium Picks: the protection idea worked, but it could be distributed much more efficiently.
The universe analyzed
The audit includes 69 closed BO3 Premium Picks between June 7 and July 27, 2026. These are positions whose main scenario was winning 2-0, expressed as a −1.5 sets handicap or correct score.
Before calculating, three voids, four still-pending matches, and one BO5 that had gotten mixed into the record were excluded. There were no missing odds, confidence values or scores within the final sample.
What changes in V4
The pick is still one per match. When protection exists, the two entries form a single Premium position: one main entry on the 2-0 and another on the match winner. The AI's selection doesn't change; the execution does.
Exposure decreases as the odds on the 2-0 scenario increase.
The main entry uses 4u in the lowest odds range, 3u in the middle range, and 2u from odds of 1.90 up. Protection only activates when winner confidence reaches 75% and its odds are at least 1.25.
When it activates, it's capped at 4u. This way, the full Premium Pick exposes between 2u and 8u and never exceeds 8% of a theoretical 100-unit bankroll.
Only the main entry is kept when the moneyline odds or confidence don't justify more exposure.
The moneyline entry adds 4u, but the total remains capped.
The same sample: V3 versus V4
To avoid misleading comparisons, both strategies were reconstructed over exactly the same 69 matches, with the same odds and results.
V4 would have produced 10.92 more units while using 119 fewer units of exposure. Maximum drawdown would also have dropped 33%, from −18u to −12u.
The improvement also shows up when splitting the timeline
- First 35 picks: +51.20u, ROI 23.49% and DD −12u.
- Last 34 picks: +32.20u, ROI 16.35% and DD −9u.
- July: +58.71u, ROI 19.38% and DD −9u.
- Last seven days available: +17.06u and ROI 19.61%.
This time split isn't a perfect future validation —the rule was designed with knowledge of the historical record— but it prevents the whole conclusion from depending solely on the first part of the sample.
What we found in the protection
The filter wasn't added because it “looked nice” in the backtest. The protections that would have passed it were compared against those that would now be left out.
The 13 entries with moneyline odds below 1.25 would have ended up practically flat: −0.12u. The two that didn't reach 75% confidence would have lost −0.47u. It's a small difference in units, but a relevant one: it avoids taking on extra risk in scenarios that didn't pay for it.
Tests to avoid falling in love with the result
A single positive backtest isn't enough. That's why we put the rule through threshold changes, odds degradation, and removal of the best results.
- 25 nearby combinations of confidence and minimum odds ended positive in both the first and second half of the sample.
- Subtracting 0.10 points from all winning odds, V4 held on to +51.30u and a 12.36% ROI.
- Removing the five best days, it would still have ended at +25.73u.
- In 50,000 daily-block resamples, the strategy ended positive in approximately 99.8% of the simulations.
That 99.8% isn't a probability of winning in the future. The resampling reuses the days already observed: it measures how much the result depends on their order and concentration, but it doesn't remove the bias of having designed the rule after analyzing the sample.
The point that still needs more data
The intermediate 3u range —odds between 1.85 and 1.89— only contains five picks. Its result was positive, but the sample is too small to claim that split is better than a binary 4u/2u formula.
The decision is to keep V4 frozen and review that range only once it accumulates at least 20 observations. Changing it now over a difference of less than one unit would be chasing noise.
Why the cap stays at 4u
A 5u or 6u protection would have earned more in this track record because the protected segment performed well. However, it would also raise the maximum loss on a single position to 9u or 10u. With a 100u reference bankroll, that would be a hard concentration to justify using only 54 historical protections.
The best formula isn't the one that squeezes out every last unit from the past. It's the one that preserves the edge without forcing us to assume the future will repeat the sample exactly.
That's why the 4u cap is kept: it seeks a balance between return, exposure and stability. V4 applies going forward to new eligible Premium Picks. Previous ones keep their original execution, and the public statistics continue to reflect what was actually delivered.
RESULTS WITHOUT REWRITING
Follow the validation in real time.
The backtest stands as a documented hypothesis. From now on, new matches will show whether the improvement holds up out of sample.
Informational content for adults over 18. Betting involves risk of loss. Past results and retrospective simulations do not guarantee future results.