Understanding Model Rarity
Models on Criesnyse are rated by rarity based on their live trading performance. This dynamic system reflects actual market results rather than theoretical potential, providing a clear measure of each model's effectiveness.
How Rarity Is Determined
Initial Rating
Every model receives an initial rarity rating after completing training and backtesting phases. This rating is based on the model's backtest performance across multiple metrics. Once deployed for live trading, the model's rarity can change based on actual trading results.
Performance Metrics
Rarity is determined by performance across six key metrics:
Return - Total gains or losses generated by the model's orders
Win Rate - Percentage of profitable orders
Sharpe Ratio - Risk-adjusted returns (higher is better)
Risk - Maximum drawdown and volatility measures
Recovery - Speed of recovering from drawdown periods
Profit Factor - Ratio of gross profits to gross losses
No single metric determines rarity. The system evaluates performance across all metrics to prevent models from optimizing for just one dimension. A model might have high returns but poor risk metrics, or high win rate but low profit factor - the rating reflects this complete picture.
Scale Adjustments
Models are categorized by the number of assets they operate simultaneously:
Single - 1 asset
Dual - 2 assets
Micro - 3-4 assets
Small - 5-19 assets
Medium - 20-49 assets
Large - 50-99 assets
Mega - 100+ assets
Performance requirements adjust based on scale category. A single-asset model faces different challenges than a mega-scale model managing 100+ assets. Larger scale models access wider trading opportunities but must manage exponentially more complex relationships. The rarity system accounts for these differences to enable fair comparison.
Rarity Tiers
Models are ranked into rarity tiers based on performance percentiles within their scale category. Rankings are dynamic - models can move up or down as their track record develops. Each tier represents a specific performance threshold relative to other models of similar scale.
Rarity Tier | Percentile | What It Means |
|---|---|---|
Common | Baseline | Most models start here. May be profitable or unprofitable - haven't yet demonstrated consistently above-average results. |
Uncommon | Top 10% | Above-average performance with consistent results across multiple metrics within scale category. |
Rare | Top 2.5% | Reliable performance across different market conditions. Maintains strong metrics even when markets shift. |
Epic | Top 1% | Consistent high performance across multiple metrics and conditions. Requires sustained excellence, not just short-term results. |
Legendary | Top 0.1% | Strong results across almost all key metrics. Very few models achieve and maintain this tier. |
Mythic | Theoretical | Near-perfect performance. No model has achieved this level yet. Aspirational category beyond Legendary. |
Rating Changes
Upgrades
Rarity upgrades are based on statistically significant improvements in performance metrics. A model must consistently outperform its current tier across multiple metrics to qualify for advancement.
Requirements increase with both tier level and scale category. Moving from Common to Uncommon requires demonstrating reliable above-average performance. Moving from Epic to Legendary requires exceptional results sustained over time. Large-scale models need stronger improvements to upgrade due to the increased complexity of managing multiple assets simultaneously.
The system looks for consistent performance, not lucky streaks. Short-term spikes in performance don't trigger upgrades - the model must maintain improved metrics over a statistically meaningful period.
Downgrades
Ratings move down as well as up. Models that experience performance decline can lose rarity status. This ensures ratings always reflect current capabilities rather than past achievements.
Models with longer successful trading histories tend to have more stable ratings. A model that's been Epic for 12 months has more established performance than one that just reached Epic last week. The longer track record reduces rating volatility because temporary fluctuations have less impact on long-term statistics.
New models experience more rating volatility as they build their track records. This is normal - early performance data carries more weight simply because there's less historical data to stabilize the statistics.
Scale Impact on Changes
Managing more assets simultaneously increases difficulty. A dual-asset model making 40% annual returns might qualify for Epic tier, while a mega-scale model would need even stronger risk-adjusted performance to reach the same tier.
This isn't arbitrary - computational complexity and market impact both grow exponentially with scale. A mega-scale model executing predictions across 100+ assets faces interconnection complexity that doesn't exist for single-asset models. The rarity system accounts for these realities.
Premium Models and Rarity
What Makes Models Premium
Premium models are trained on three or more assets simultaneously. This allows them to capture complex relationships between assets - for example, how currency movements affect commodity prices, or how sector rotation impacts individual stocks.
Training premium models requires exponentially more computational resources than basic models (1-2 assets). The complexity doesn't scale linearly - adding a third asset doesn't just add 50% more work, it multiplies the computational requirements. Each additional asset increases the number of relationships the model must learn, creating exponential growth in both training time and resource needs.
Computational Costs
Training and backtesting consume the most computational resources. Training a 5-asset model can require 10x the GPU hours of a 2-asset model. Backtesting similarly scales up as the model must process predictions for all tracked assets across years of historical data.
Live operation also requires more resources for premium models, though less dramatically. Generating daily predictions for multiple assets simultaneously still demands more computing power than single-asset predictions, which is why premium model access exists as a separate subscription tier.
Rarity Considerations
Premium models don't automatically receive higher rarity. A poorly-performing 10-asset model can be Common tier, while an exceptional single-asset model can be Legendary. Performance determines rarity, not scale.
However, premium models compete within appropriate scale categories. A Small-scale model (5-19 assets) is evaluated against other Small-scale models, not against Single-asset models. This ensures fair comparison - you can't directly compare the difficulty of managing 1 asset versus 50 assets.
Premium models that achieve high rarity tiers demonstrate mastery of complex multi-asset relationships while maintaining strong risk-adjusted performance. This combination of scale and quality makes high-tier premium models relatively rare in the ecosystem.
Key Takeaways
Performance-based ratings - Rarity reflects actual trading results, not theoretical potential or marketing claims. Every model's rating is based on verified predictions and measurable outcomes.
Dynamic system - Ratings update based on live performance. Yesterday's Legendary model can become tomorrow's Epic model if performance declines. This ensures ratings remain current and meaningful.
Fair comparison - Scale adjustments enable fair comparison between models of different complexity levels. Single-asset and mega-scale models are evaluated appropriately for their respective challenges.
Time validates ratings - Longer track records create more reliable rarity assessments. A model maintaining Epic status for 12 months has more proven capability than one that just reached Epic tier.
Premium means complexity, not quality - Premium models analyze more assets simultaneously but aren't inherently better. Some single-asset models outperform complex multi-asset models. Choose models based on performance and your needs, not just scale.
FAQ
Can rarity change after deployment?
Yes. Rarity updates continuously based on live trading performance. Models can move up or down in rating as their track records develop.
Do premium models automatically get higher rarity?
No. Premium models aren't rated higher simply for analyzing more assets. Performance determines rarity regardless of model scale. A well-performing single-asset model can be Legendary while a poorly-performing 20-asset model remains Common.
How long does it take for rarity to stabilize?
This varies by trading frequency and consistency. Models that trade daily build statistical significance faster than models that trade weekly. Generally, several months of live trading data is needed before rarity becomes stable. Models with longer track records experience less rating volatility.
What's the difference between initial and live rarity?
Initial rarity is based on backtest performance - how the model performed on historical data. Live rarity reflects actual trading results after deployment. The two can diverge as models encounter market conditions different from their training period.
Can a model lose rarity status?
Yes. If a model's live performance declines, its rarity can drop. Ratings always reflect current performance, not past achievements. A model that was Rare based on strong backtest results can become Uncommon or Common if live trading results don't maintain that level.
Does higher rarity mean I should track that model?
Not necessarily. Rarity indicates past performance, but doesn't guarantee future results. Consider multiple factors: the model's track record length, performance across different market conditions, whether it trades assets you're interested in, and how its classification style matches your trading approach. Higher rarity is meaningful, but it's one factor among many.