network-quality-scoring

NetworkQualityMonitor Scoring Algorithm

Overview

NetworkQualityMonitor uses a browser-optimized scoring algorithm that prioritizes latency and consistency over raw bandwidth. The scoring is weighted as follows:

• HTTP Response (40%): Most critical for browser experience
• Bandwidth (35%): Important for media and downloads
• DNS (10%): Foundation of all connections
• Buffer Bloat (15%): Network congestion under load

HTTP Response Scoring (40% Weight)

Measurement Process

1. Endpoints: Tests 10+ global endpoints including DuckDuckGo, major CDNs, and popular platforms
2. Sampling: 15 requests per endpoint with interleaved ordering to avoid bias
3. Statistical Analysis:
   • Calculates median response time for each site
   • Uses median of all site medians for overall response time
   • Calculates standard deviation for each site's measurements
   • Uses median of all site standard deviations for consistency metric

Response Time Calculation

// For each site: calculate median of its measurements
let siteMedians = sites.map { calculateMedian($0.measurements) }

// Overall response time: median of all site medians
let overallResponseTime = median(siteMedians)

This approach:

• Resists outliers at both per-site and cross-site levels
• Provides typical latency experience across geographic regions
• Doesn't get skewed by one particularly fast CDN or slow server

Consistency Calculation

// For each site: calculate standard deviation of its measurements
let siteStdDevs = sites.map { calculateStdDev($0.measurements) }

// Overall consistency: median of all site standard deviations
let overallStdDev = median(siteStdDevs)

This avoids mixing different latency populations (e.g., 20ms CDN vs 300ms cross-ocean).

Coefficient of Variation (CV) Scoring

The scoring uses Coefficient of Variation to fairly compare different latency ranges:

// CV = stdDev / mean - normalizes variance relative to baseline
let coefficientOfVariation = stdDev / averageResponseTime

// Apply CV-based penalty (fairer than raw standard deviation)
// Example: 10ms variance on 50ms latency (20% CV) penalized more than
//         10ms variance on 200ms latency (5% CV)