ArcaThread desktop app is still in development and coming soon.

How It Works

From Data to Decision Support

ArcaThread combines LCU game-state capture, patch-aware logic, and ML-assisted scoring to surface actionable suggestions.

Data Sources

Claims on this page map to active code paths in the desktop app and runtime services.

Official Client Data

Runtime state is read from official League Client (LCU) endpoints and related aggregation services.

Patch-Aware Routing

Recommendation logic is patch-aware and includes fallback paths when coverage is sparse.

Role and Rank Context

Draft/live scoring can include role assumptions and rank-aware model routing.

Region-Aware Inputs

Current runtime focuses on supported regions in your configured environment.

Recommendation Pipeline

The runtime flow below reflects active application architecture.

01

Game Detection

ArcaThread detects champion select and live games through LCU events and polling.

  • LCU event-driven updates
  • Draft/live phase recognition
  • No game process memory access
02

Context Extraction

The app builds a normalized state payload from champions, items, objectives, and timing context.

  • Patch and queue context
  • Team composition extraction
  • Gold, level, and item state features
03

Scoring and Retrieval

Context is routed through stats-backed and ML-assisted scoring paths with deterministic fallback behavior.

  • Counter-item analysis categories
  • Rank-aware model lookup
  • Fallback chain when data/model is unavailable
04

ML Inference Layer

ONNX runtime support is used for local inference paths, alongside fallback logic to keep outputs available.

  • ONNX runner integration
  • Model service caching
  • Graceful fallback to non-ML recommendation paths
05

Suggestion Output

Recommendations are presented as options with rationale. The player always makes the final decision.

  • Suggestion-only UX
  • Build path and alternative options
  • Configurable interval (30s default)
Adaptive Logic

Live Context Signals

Live recommendations can react to multiple state signals, not only static matchup tables.

Game Phase

Recommendations can differ between lane, mid game, and late game context.

Enemy Itemization

Counter-item categories include anti-heal, defenses, penetration, and utility responses.

Timing and Tempo

Recommendation refresh follows runtime intervals with a 30-second default setting.

Embedding Layer

Champion features include 20-dimensional embeddings used in multiple recommendation paths.

We value your privacy

We use cookies for analytics and optional ad conversion measurement. You can accept all, reject all, or customize your choices.