Changelog
Changelog
Update Log & Improvements
Changelog:
- Released: v10_unified — completely retrained AI model on a unified, expanded dataset of 233,682 verified CS2 profiles
- Enhanced: F1-Score improved from 86.29% to 87.83% (+1.54), reaching the highest precision/recall balance achieved on the platform
- Boosted: ROC-AUC raised from 93.60% to 95.40% (+1.80), confirming exceptional class-separation power
- Improved: Recall (cheater catch rate) up from 88.75% to 91.15% (+2.40), catching more real cheaters at the same threshold
- Refined: Precision lifted from 83.97% to 84.75% (+0.78), reducing false-positive risk
- Achieved: 89.34% overall accuracy on the held-out test set
- Expanded: Feature space grew from 180 to 350 selected features, including new cross-source consistency signals (Leetify ↔ Scope ↔ Steam) and 12 "broken-pattern" anomaly detectors
- Re-architected: Switched from a single RandomForest to a 6-model ensemble — RF balanced, ExtraTrees balanced, XGBoost balanced, XGBoost conservative, LightGBM fast, LightGBM precise — combined with weighted soft voting
- Tuned: F1-optimal threshold validated at 0.52, decision threshold kept at 0.5 for backward compatibility
- Visual: New live metrics widget on the analyzer page showing real-time deployment status of v10_unified
Changelog:
- Added: Comprehensive Leetify data integration with granular analysis of aim metrics, positioning, utility usage, and consistency patterns - enabling detection even with private Steam profiles
- Enhanced: Visual reasoning system with responses showing specific suspicious patterns detected, confidence levels, and data sources used for each analysis
- Improved: Private profile handling - system now provides predictions when sufficient alternative data (Leetify) is available, reducing "no data" responses by 73%
- Optimized: Pattern detection algorithms to identify specific cheating behaviors including reaction time anomalies, unnatural consistency scores, and sudden skill improvements
- Added: Multi-tier confidence assessment based on available data sources, clearly communicating prediction reliability to users
- Upgraded: Response generation with explanations featuring color coded risk boxes, visual data source indicators, and detailed reasoning for each prediction
- Boosted: User trust through transparent AI decision-making, showing exactly why the system flagged or cleared each player profile
Changelog:
- Added: Full Leetify API integration providing access to 30+ advanced performance metrics including T-side opening aggression success (40.18% importance), counter-strafing ratio, and reaction time analysis
- Enhanced: Detection accuracy improved from 81.98% to 86.29% F1-Score by incorporating Leetify's professional statistics - specifically clutch ratings, map-specific performance, and Premier/Faceit rankings
- Improved: False positive reduction from 4,543 to 3,258 cases (-28%) through Leetify's verified match data validation and opening duel success rate analysis
- Optimized: Feature importance rebalancing - Leetify metrics now represent 7 of top 20 most important features, with T-side aggression becoming the #1 predictor
- Fixed: Cross-game ban contamination by validating CS2 activity through Leetify's match history before classification
- Upgraded: ROC-AUC performance boost from 89.92% to 93.60% by combining internal behavioral analysis with Leetify's competitive statistics
- Boosted: High-confidence predictions increased from 47.9% to 70.5% of all analyses through Leetify data enrichment
Changelog:
Fixed: rare cases where players banned in other games (e.g. MW2, Dota 2, TF2,...) were being treated as CS2 cheaters, polluting the model with false positives
Added: a simple CS2 activity check anyone with "X" minutes of CS2 play in the last 2 weeks is now excluded from the “cheater” group if their ban comes from another title
Improved: training-data purity by removing these unrelated bans, so genuine CS2 player behavior drives the model’s learning
Boosted: overall accuracy and confidence cheaters and legitimate players are now far better separated, with far fewer false alarms
Changelog:
Upgraded: Core AI model retrained with expanded dataset (+202K profiles analyzed) for enhanced detection precision
Enhanced: Accuracy of cheater detection through improved gameplay feature extraction and refined inventory analysis
Optimized: Training pipeline to prioritize reliable behavioral patterns over noise, significantly reducing false positives
Added: Advanced derived gameplay statistics (K/D ratio, headshot accuracy, win ratio) for better decision-making
Improved: Inventory parsing to better identify rare skins and suspicious asset patterns
Optimized: Feature selection process to ensure only the most valuable 70 features contribute to final decision-making
Boosted: Overall model confidence and performance for production-level deployment
Changelog:
- Added: Progressive Web App (PWA) support to allow installation as a standalone app
- Enhanced: Navigation experience with faster load times and offline access
- Added: Automated creation of player profile pages after each analysis
- Improved: Dynamic linking to profiles using Steam ID data
- Upgraded: Sharing capabilities with direct access to analyzed profiles
- Optimized: Rendering flow for smoother transitions between pages
Changelog:
- Added: New tracking feature to record how many times a profile has been analyzed
Changelog:
- Fixed: Game ban detection issue that was missing certain banned accounts
- Improved: Ban tracking system to combine VAC and Game bans for more accurate cheater detection
- Enhanced: Database recording to properly flag accounts with any type of ban (VAC or Game)
- Added: Unified ban count that includes both VAC and Game bans in the NumberOfVACBans field
- Upgraded: Friend ban detection to identify friends with either VAC or Game bans
- Optimized: Ban data collection to adapt to Valve's evolving ban system implementation
Changelog:
- Fixed: Menu positioning on mobile and tablet devices
- Enhanced: Z-index handling for better element layering
- Improved: Button visibility management during menu navigation
- Optimized: Mobile interface responsiveness across different screen sizes
- Added: Smart element hiding for cleaner user experience
Changelog:
- Improved: Resource rotation system for more efficient service utilization
- Enhanced: Response times by eliminating unnecessary retry attempts
- Fixed: Performance bottlenecks when dealing with private profiles
- Added: Smart error handling for better resource management
- Optimized: Overall processing speed for player data collection
Changelog:
- Fixed: Issue where simultaneous requests received identical results
- Added: Better handling of concurrent analysis tasks
- Improved: Request data isolation for multiple users