Composite Rebuzz Score

What it measures

The Rebuzz score is the final "single number" summary of the model.

It combines four different views of market conversation:

• Agreement across communities (CCSC).
• Elite vs crowd disagreement (CDS).
• Timing and spread quality (TCIS).
• Sudden baseline breaks (AWVA).

The output includes both:

• A directional score.
• A confidence value used to gate labels.

Inputs

The composite takes the latest outputs from CCSC, CDS, TCIS, and AWVA.

Some components can be missing (for example, AWVA during cold-start history). The model is built to handle that.

How it is computed

1. Put all component scores onto a comparable scale.
2. Apply base weights:
   • CCSC: 30%
   • CDS: 25%
   • TCIS: 22.5%
   • AWVA: 22.5%
3. If a component is missing, redistribute its weight proportionally across available components.
4. Compute weighted average.
5. Attach confidence:
   • Use CCSC confidence when available.
   • Otherwise use a fallback based on how many components were available.

If fewer than two components are available, Rebuzz returns INSUFFICIENT_DATA instead of forcing a weak signal.

Output + interpretation

Score range is -1 to +1:

• Positive = bullish tilt.
• Negative = bearish tilt.

Labels are confidence-gated, not score-only:

• STRONG_BUY: score > 0.7 and confidence > 0.7.
• BUY: score > 0.4 and confidence > 0.5.
• SELL: score < -0.4 and confidence > 0.5.
• STRONG_SELL: score < -0.7 and confidence > 0.7.
• Otherwise: HOLD.

Edge cases/default behavior

The composite is designed to stay stable when data availability changes day to day.

• Missing components are handled by weight redistribution.
• Confidence is always bounded to 0 to 1.
• Too little information returns INSUFFICIENT_DATA.

How Rebuzz uses it

This is the score used across Rebuzz experiences such as rankings, ticker pages, and alert decisions.

It is recomputed daily for active tickers and stored with both label and confidence for transparency.