← 📡 FableRadar

Methodology

How the score is computed.

Signals

Three weighted signals, each scored 0-100 (50 = neutral), refreshed every 30 minutes from free public sources:

  • Anthropic Status (25%) - distinct articles on anthropic.com/news classified for restore- vs hold-language. Floored at 35 so a single all-negative source can't zero it.
  • Regulatory Climate (35%) - Google News RSS on the export directive, scoring easing vs tightening language (word-boundary matched, smoothed on thin samples).
  • Prediction Markets (25%) - volume-weighted YES odds across open Manifold markets on Anthropic restoring access to Fable 5. The crowd's aggregated forecast.
  • Community Buzz (15%) - volume of Hacker News discussion over the last 7 days (log-scaled).

Composite

The signals are combined as a weighted mean. When two or more signals are simultaneously "hot" (above 60), a bounded alignment bonus closes part of the gap to 100 (it is additive, not a multiplier, to avoid saturation).

Smoothing

The composite is passed through an asymmetric exponential moving average: it rises quickly on good news and decays slowly (a built-in skepticism). Bands: Frozen 0-30, Uncertain 31-60, Thawing 61-85, Return Imminent 86-100.

A note on the trend chart

History before launch day is a model reconstruction - the real scoring engine run over plausible past signal values from the June 12 suspension - not measured readings. Everything from launch onward is live.