← Back to the Divergence Index
How the Divergence Index is computed
Where prediction markets disagree with the press. For each event Storyflo holds both a liquid prediction-market contract and a set of narrated news stories, we compute the gap between the market's implied probability and the probability the news narrative implies for the same event question.
The metric
For each matched event we compute two probabilities on the same [0, 1] scale for the same YES/NO question, then take the gap:
divergence = market_implied_probability - news_narrative_probability
Events are ranked by abs_divergence = |divergence| — the absolute divergence. Both probabilities are on a [0, 1] scale for the YES outcome of the matched market question.
- market_ahead
- Positive divergence — the market is more confident in YES than the press narrative.
- press_ahead
- Negative divergence — the press narrative leans YES harder than the market is pricing.
- aligned
- Within +/- 0.02 — market and press agree.
Magnitude bands — mild (< 0.15), notable (0.15 - 0.30), stark (>= 0.30).
The news-narrative probability (Storyflo's own number)
A calibrated language-model pass reads the matched stories' headlines and summaries and estimates the probability the coverage implies for the YES resolution of the market question. It is a read of the news narrative's implied confidence — not a Storyflo forecast and not a market price.
Source: Computed by Storyflo (original analysis).
The market-implied probability (an editorial input)
External prediction markets (Kalshi, Polymarket), used as an editorial INPUT.
Storyflo publishes its OWN divergence and news-narrative numbers. The underlying raw market probability is withheld by default (it is the venue's price feed, not Storyflo's analysis); it is shown only when Storyflo enables it.
Matching stories to markets
Stories are matched to market contracts with a precision-first, document-frequency-weighted token matcher (a link requires >=3 shared distinctive tokens, or 2 with one rare token). False positives are treated as worse than false negatives.
Cadence
Recomputed several times daily; published as a weekly index.
