Ten strategies.
The same recorded outcomes.

01 / Shared starting conditions

Ten rules, eight separate sessions.

Each recorded session receives a fresh hypothetical 1,000-credit bank. Base stake: 4 credits; stake cap: 64. The percentage rule instead starts at 1% of the bank. Limits are −100 or +100 credits; the blank-streak threshold is three and the interval threshold is five. These are ten selected, explicitly defined rules, not a measured popularity ranking.

A payout below the stake is a net loss. Every rule pays for every inspected spin, including the blank spins used by the two-level rule. Changing a stake does not change the recorded payout multiplier in this mathematical model.

Drawdown is the largest fall from a previous bank peak within any one replayed session. It is not the final loss, and the eight drawdowns are not added together.

02 / Fitted to the entire known sample

The initial 660 settings: no profitable complete replay.

We tried base stakes of 1, 2, 4, 8 and 16 credits and caps of 32, 64, 128 and 256. Percentage settings were 0.25%, 0.5%, 1%, 2% and 4%; blank-streak and interval thresholds were 2, 3, 5, 8 and 12. Session limits combined loss and profit thresholds of 50, 100, 200 and 400. Duplicate percentage configurations with an unused base stake were excluded.

We rank only settings that inspect all 699 rows across all eight sessions. A positive early exit is not a profitable full-sample run. Searching an already-known sample is fitting, not an unseen test. Different banks, grids or samples can change the result.

Practical takeaway: a smaller stake can make a loss smaller. It does not turn the result into profit. There is no cashout multiplier to tune in this slot, and raising a stake after blanks does not guarantee that the next spin will pay.

03 / Why a short result can mislead

Earlier rows select; later rows replay.

Within each of the eight sessions, the first 60% of rows select settings; the remaining rows replay those settings with a new 1,000-credit bank. That gives 418 early and 281 later rows. The dataset was already known, so this is a retrospective sensitivity check, not independent validation. Short sessions yield very small splits.

Selection prefers a configuration completing all early slices. If none does, the largest early net is shown as an incomplete selection. Always read completed slices and bets beside profit: stopping after a lucky payout can make a short run look impressive.

The interval-doubling selection went from +1,183.52 credits in the early slices to −663.37 in later slices. Splitting and restarting a progression changes the path, so these figures must not be added to claim a continuous full-sample profit.

04 / Try the arithmetic

Your settings. One recorded session.

This calculator replays existing payout multiples; it does not launch the game, place bets or predict outcomes. Choose a session to keep its real boundaries intact. Amounts are hypothetical credits, not ZAR or verified live stake options.

Percentage, streak, interval and stop-limit inputs apply only to their named rules. “Full-sample best” means the least-loss complete replay in this grid, not a recommended strategy. Loading it preserves your chosen starting bank; changing that bank changes the result.

Choose a rule and session, then run the replay.

The line shows hypothetical bank credits from left to right. Top label: the chart maximum; bottom labels: recorded row positions. An early endpoint means a cap, balance or session limit stopped the replay.

05 / Reproducible research

What we ran, and what we did not.

Each session starts a fresh hypothetical 1,000-credit bank and fresh progression state. Aggregate net is the sum of eight independent replays, not one uninterrupted balance. Recorded payout/stake multiples are scaled linearly to hypothetical stakes; actual stake-dependent payouts and denominations have not been tested. No new spins were collected. Partial sample: 699 accepted, 27 excluded after cutoff; unresolved requests and mixed transport are documented in Game Lab.

Source observations were collected on 13 October 2026: 530 through the ordinary interface and 169 through sequential public-demo requests. The original stake was 4 demo credits. The observed return was 2,620 / 2,796 = 93.71%; this is not theoretical RTP. No new live spins were collected for this strategy comparison.

Each payout is scaled by its recorded payout-to-stake ratio, then rounded half-up to the nearest cent. The model does not establish that every fractional stake is available or that changing coin value and bet level preserves every paytable award. Blank-streak calculations pay for their observation spins. Progressions reset at recorded session boundaries. A complete replay must inspect every row in every session; the eight fresh banks represent 8,000 hypothetical credits in total.

Net profit = total returned minus total staked. Drawdown = largest fall from an earlier bank peak. Partial payouts still count as losses for loss-based progressions; Paroli and 1–3–2–6 advance only on net profits. A stake exceeding the cap or bank is never placed. No automatic reset hides a failed session.

Read the original collection method and missing outcomes →

Save the JSON outcomes, results, search CSV, engine and reproduction script in one folder. Run node reproduce-strategies.mjs using Node.js 18 or later; no packages or network access are required. The script reproduces every baseline, all 660 grid configurations and the published full-sample selections, not the original random outcomes.