<div class="section-label"><span class="num">04</span> MaxFit</div>
<div class="section-title">Search thresholds for the strongest <em>fit</em>.</div>
MaxFit keeps a **fixed** attribute set and searches candidate thresholds (optionally both relevance and similarity censoring), then returns the single winning prediction. You choose the objective (fit, adjusted fit, or k-fit); the engine does the search.
It is a useful building block when you want threshold selection without a full attribute grid. For primary analysis, prefer [[Grid Prediction]], which diversifies across attribute combinations *and* thresholds, then blends by fit.
<p> </p>
## When to use MaxFit
- You want the engine to **pick** the threshold on one attribute set
- You may want to compare relevance and similarity censoring in one call
- You care about an explicit objective: <span class="mono">fit</span>, <span class="mono">adjusted_fit</span>, or <span class="mono">kfit</span>
> [!tip]
> Need attribute combinations, a fit-weighted composite, and variable impact measures? That is [[Grid Prediction]], the recommended default. MaxFit is the right tool for threshold-focused work on a set you already chose.
> [!note]
> Fixed thresholds you already know → [[Predict]]. Exploring many attribute subsets → [[Grid Prediction]].
## What you call
```text
predict_maxfit(
y, # outcomes, length N
X, # attributes, N × K
theta, # circumstances, length K
options? # optional; omit → defaults
) → prediction results
```
Same data shape as [[Predict]].
## Settings that matter
**MaxFit-only**
| Setting | Meaning | Default |
| --- | --- | --- |
| <span class="mono">objective</span> | Score used to pick the winner | <span class="mono">kfit</span> |
| <span class="mono">inner_parallel</span> | Speed up searching both censor types | <span class="mono">auto</span> |
**Shared settings** (see [[Config/Options#MaxFitOptions|MaxFitOptions]])
- Default thresholds: <span class="mono">[0, 0.2, 0.5, 0.8]</span>
- Default censor type: <span class="mono">both</span> (allowed here; not on Predict)
- Plus the usual censor unit / operator, scale, and flags
## What you get back
**PredictionResults** for the **winning** threshold: typically one prediction, not a combination grid.
- Root fields such as <span class="mono">yhat</span> and <span class="mono">fit</span> reflect the winner ([[Results/Core|Core]])
- <span class="mono">maxfit_index</span> (when present) marks which candidates were in the original search
- Nested groups (<span class="mono">prediction_weights</span>, <span class="mono">insights</span>, <span class="mono">auxiliary</span>, <span class="mono">solo_distribution</span>) follow the same pattern as Predict for the retained result
<div class="btn-row center">
<a class="btn-primary" href="/Functions/Grid%20Prediction">Prefer Grid Prediction</a>
<a class="btn-ghost" href="/Config/Options#MaxFitOptions">MaxFitOptions</a>
</div>
## Related
- [[Grid Prediction]]: primary analysis path (recommended)
- [[Predict]]: single evaluation at given thresholds
- [[Results/Overview|Results]]: how to read the winning result