<div class="section-label"><span class="num">03</span> Weights & Insights</div>
<div class="section-title">Which observations mattered, and <em>why</em>.</div>
## Observation weights
Observation weights describe which observations mattered and whether they were included or excluded in forming the prediction.
| Field | Role |
| --- | --- |
| <span class="mono">include</span> | Which observations entered the forecast at each threshold |
| <span class="mono">weights</span> | Weights on the included side |
| <span class="mono">weights_excluded</span> | Weights on the excluded side |
Typically shaped N × T. Example: <span class="mono">result.prediction_weights.weights</span>
## Scores
The same family as standalone [[Relevance Metrics]], computed as part of the prediction.
| Field | Shape | Role |
| --- | --- | --- |
| <span class="mono">relevance</span> | N | Relevance of each training row to <span class="mono">theta</span> |
| <span class="mono">similarity</span> | N | Similarity of each row to <span class="mono">theta</span> |
| <span class="mono">info_x</span> | N | How unusual each row is vs the sample mean |
| <span class="mono">info_theta</span> | 1 | How unusual the circumstances are |
| <span class="mono">weights_concentration</span> | T | How concentrated the prediction weights are |
Example: <span class="mono">result.insights.relevance</span>
## When they appear
| Path | Typical |
| --- | --- |
| Predict / MaxFit | Present |
| Grid | Usually missing; use [[Results/Grid Insights\|Grid Insights]] |
A group that was not filled is simply missing from the result.
## Related
- [[Settings/Options#PredictOptions|PredictOptions]]: <span class="mono">censor_type</span> chooses the score that drives include vs exclude
- [[Results/Core|Core]]: the forecast and fit
<div class="btn-row center">
<a class="btn-primary" href="/Results/Solo%20Distribution">Solo Distribution</a>
<a class="btn-ghost" href="/Functions/Relevance%20Metrics">Relevance Metrics</a>
</div>