Source Reliability Methodology

Source Credibility Weighting

btc.london rates every news source on a 0.55–1.0 credibility scale based on four factors: editorial independence (does the outlet have a financial stake in crypto?), fact-checking standards, track record for accuracy, and transparency about ownership and funding.

Why this matters: one Reuters article outweighs five Bitcoin Magazine articles on the same claim. Without source weighting, a coordinated PR push from a single project can temporarily dominate a naive aggregator's sentiment reading.

Source Registry (selected)

Bias Label Assignment

Bias labels are assigned using source-weighted keyword scoring. Keywords associated with bullish narratives (adoption, ATH, surge, institutional, breakthrough, record) increase the Pro-Crypto score. Bearish keywords (crash, ban, fraud, collapse, warning, lawsuit, hack) increase the Skeptical score. Keywords appearing in the headline receive 1.5× the weight of keywords in the article body.

Thresholds: weighted score above +0.15 = Pro-Crypto; below -0.15 = Skeptical; between -0.15 and +0.15 = Neutral. Claude Sonnet provides an additional narrative layer (20% of the final coin sentiment blend) that catches tonal nuances keyword matching misses.

Feed & Deduplication

• Refresh cycle: Every 30 minutes, ~250 articles per cycle from 15+ RSS feeds
• Deduplication: Jaccard similarity at 0.62 threshold on normalised title tokens — duplicate stories from multiple outlets are collapsed to the highest-credibility source
• Image backfill: Articles without images are backfilled from Open Graph metadata of the source URL
• Category detection: Articles are automatically categorised (Bitcoin, Ethereum, DeFi, Regulation, etc.) based on keyword matching against per-category keyword sets
• Pre-curation: Low-quality noise sources, content farms, and purely promotional outlets are excluded from the feed before any analysis runs