From Data to Decisions
ArcaThread combines patch-aware statistics, machine learning, and real-time game state to provide recommendations that actually make sense.
Data Sources
We don't guess. Every recommendation is backed by real match data, segmented by the factors that actually matter.
Riot API Data
Millions of ranked matches from Riot's official API, continuously ingested and processed.
Patch Versioning
Every recommendation is tied to a specific patch. No stale data from pre-rework champions.
Role & Rank Context
Diamond support builds differ from Silver. We segment data by role and rank for precise recommendations.
Regional Meta
Currently supporting EUW, with more regions planned. Our models will account for regional differences as we expand.
The Recommendation Pipeline
From the moment you enter a game to your final item purchase, here's what happens behind the scenes.
Game Detection
ArcaThread detects when you enter champion select or an active game via the League Client API (LCU). No manual input required.
- LCU WebSocket connection for real-time events
- Automatic draft phase recognition
- Live game state monitoring
Context Extraction
We gather all relevant context: your champion, role, enemies picked, current patch, and your rank/region.
- Champion select anonymity preserved (ally 1-5)
- Enemy team composition analysis
- Patch-aware feature derivation
Data Retrieval
Query our aggregated statistics for the exact matchup context: win rates, item efficiencies, rune synergies.
- Sub-100ms query response time
- Fallback chain: exact → rank backoff → patch backoff
- Global safety net for new patches
ML Inference
Our ONNX models rank options based on millions of match outcomes in similar contexts.
- Local inference for speed (<50ms)
- Server-side for complex models when needed
- Deterministic fallback if models unavailable
Recommendation
Present 2-3 viable options with clear rationale. You decide—not an autopilot.
- Build path visualization
- Situational reasoning included
- Updates every ~30 seconds in-game
Recommendations That Adapt to the Game
Static build guides fail because they don't account for the dynamic nature of League matches. ArcaThread reads game state and adjusts in real time.
Game Phase
Early game priorities differ from late game needs
Enemy Builds
Detects enemy itemization and suggests counter-builds
Gold Differential
Behind? Safer items. Ahead? Push your advantage.
Team Composition
Considers your team's damage types and gaps
Live adaptive recommendations view
Real screenshot coming soon
Built for Performance
<100ms
Query Latency
P95 response time
<50ms
Model Inference
Local ONNX runtime
~30s
Update Interval
In-game recommendations
Low
Memory Usage
Lightweight footprint