Skip to main content
Previa’s scoring engine runs every 10 minutes across all active markets. It combines five data signals into a single 0-100 composite score.

​
The five scoring components

​
1. Market price signal (40% weight)

The largest component. The scoring engine analyzes:
  • Current price — Where the YES/NO contract is trading right now
  • Price momentum — Direction and velocity of recent price movement
  • Price trajectory — Whether the trend is accelerating, decelerating, or reversing
This captures what the market’s collective intelligence is pricing in.

​
2. News signal (25% weight)

Previa ingests news articles from multiple sources and maps them to markets using AI:
  1. Ingestion — Articles are fetched from news providers every 15 minutes, deduplicated, and stored.
  2. Analysis — Each article is analyzed by an AI model (Claude) for topic, sentiment, and potential market impact.
  3. Matching — Articles are matched to relevant markets using semantic similarity (OpenAI embeddings). An article must exceed a 0.7 similarity threshold to be mapped.
  4. Scoring — The aggregate sentiment and impact of matched articles feeds into the market’s score.

​
3. Social sentiment (15% weight)

The social intelligence layer is being expanded. Sentiment data is currently available for markets with active social discussion. Coverage will grow as more sources are integrated.
Social sentiment is gathered from X (Twitter), Reddit, and Telegram. The engine calculates:
  • Weighted sentiment — Each post is scored for sentiment, weighted by the source’s credibility
  • Rolling windows — Sentiment is aggregated over 1-hour, 6-hour, and 24-hour windows
  • Anomaly detection — Sudden sentiment shifts or volume spikes are flagged

​
4. Historical patterns (15% weight)

For markets with enough historical data, the engine compares:
  • Similar resolved markets — How markets with comparable characteristics resolved
  • Category trends — How markets in the same category tend to behave
This component strengthens over time as more markets resolve and the dataset grows.

​
5. Time decay (5% weight)

As a market approaches its resolution date, the time decay factor increases. This reflects the narrowing window for price movement and the increasing weight of existing evidence.

​
Score calculation

Each component produces a sub-score from 0 to 100. The composite score is the weighted sum: 
composite = (price × 0.40) + (news × 0.25) + (social × 0.15) + (historical × 0.15) + (decay × 0.05)
The result is rounded to the nearest integer.

​
AI reasoning

When a market’s score changes by 5 or more points between cycles, the scoring engine generates a natural-language explanation using Claude. This reasoning:
  • Identifies which component(s) drove the change
  • Summarizes the key evidence (news articles, price movements, sentiment shifts)
  • Provides context for why the score moved
Reasoning is available to Pro and Premium users on the market detail page.

​
Score alerts

When a score changes by more than 10 points in a single cycle, a score alert is generated. You can subscribe to these alerts for specific markets. See alert types.

​
Limitations

  • Scores reflect the data available to Previa at scoring time. They do not account for insider information or events that haven’t been reported.
  • The social component depends on public discussion volume. Low-profile markets may have limited social signal.
  • Historical comparisons are strongest for categories with many resolved markets (politics, economics) and weakest for novel or one-off events.
  • Scores are analytical signals, not trade recommendations.