Method
How a model earns its release
DATA → CONDITIONS → SIGNAL → VALIDATION → DELIVERY. Only ACTIVE models distribute actionable signals. Reaching 200 signals makes a model eligible for review — it never activates a model automatically.
Validation gates
Sample size
Completed valid signals
Profitability
ROI across the full validation sample
Controlled losses
Maximum drawdown, standardized 1-unit staking
Consistency over time
First 200 signals in four consecutive blocks of 50
No jackpot dependence
Stress-test whether a few outsized winners explain the result
Diversification
Competition and team concentration
Closing line value
CLV against the approved fair-price close
The +3.5% ROI figure is a qualification threshold, not a promise of future ROI. The 15-unit drawdown limit and the outlier stress-test are initial working rules and will be recalibrated once enough real model histories exist.
The decision at 200 signals
Pass all required gates
The model becomes ACTIVE after human approval. Telegram is enabled and the full model profile is published.
Fail with a diagnosed weakness
The rules are modified, a new version is created and the counter resets to zero for a fresh 200-signal cycle.
Fail with no defensible fix
The model is rejected or retired and replaced by a new model concept.
Performance from different model versions is never mixed. A reset means a new validation sample from zero.
Model states
Collecting real signals and building a track record.
Minimum sample reached; all gates are being checked.
Passed validation and approved for users.
Rules changed; the new version restarts from zero.
Failed with no justified improvement path.
AI Data Audit
Don't just trust MODIX. Challenge it.
Give our audit summary to your own AI and let it check us. We publish how we test, what the checks found, and where the limits are.
Methodology
How bets are settled, how returns and CLV are calculated, how seasons are used for discovery and consistency.
Check results
What our data checks found and which matches were excluded, and why.
Limitations
What the data cannot tell you yet — plainly stated.
Raw bookmaker prices are not published, because of data-provider restrictions.
Show the ingredients. Protect the recipe.
Public: the types of inputs each model reads — xG, shots, momentum, score, minute, corners, form and home/away data. Private: thresholds, weights, formulas, rule combinations and proprietary filtering logic.
How performance is reported
Every published figure states the period, sample size, odds basis, staking method (standardized 1 unit) and whether the results are pre-release validation or live/public performance. No fabricated records, no cherry-picking, no hypothetical profits.
Statistical models, not AI characters
MODIX90 models are digital tipsters: rule-based, data-driven statistical models. AI is not used as a marketing label.
Human approval is the final step
Metrics are calculated automatically from stored signals, but qualification never silently switches a model to ACTIVE.