The Prediction Profiler lets you explore how the model's predictions respond as you change one factor at a time, around a point you choose — without running any new experiments. It is a fast way to build intuition for what the model has learned: which factors move an objective the most, where a response flattens out, and how confident the model is across each factor's range.
Where to Find It
Tab: Overview → Predictions section.
Each row in the predictions table has a Profile button next to Visualize. It becomes available once you have generated a prediction for that row and the model is trained and up to date.
How to Use It
Generate a prediction for a row in the predictions table, then click Profile on that row.
Your anchor point is that row. The profiler opens with every factor held at the row's values — this is the point the cross-sections pass through. You can edit the anchor directly in the center-point row at the top of the profiler; predictions update as you change it.
Choose which factors to sweep with the X Axis selector, and which objectives to show with the Y Axis selector.
Reading the Charts
The profiler is a grid of small panels: one row per objective, one column per swept factor. Each panel shows how that objective's prediction changes as the factor varies, with every other factor held at the anchor.
Numerical factors — a line traces the predicted value across the factor's range, with a shaded band showing the model's uncertainty (±1 standard deviation). A marker shows the anchor value on that factor.
Categorical factors — one point per category, each with an uncertainty bar, shown without a connecting line.
Panels in the same objective row share a y-axis, so you can compare the effect of different factors at a glance.
Click a point on a line (or a category) to move the anchor to that value and re-profile around the new point.
Good to Know
It reads the model, it does not run experiments. Everything you see is the surrogate model's prediction — no new measurements are consumed.
One factor at a time. Each panel varies a single factor and holds the rest at the anchor, so it shows local behavior around that point — not the global optimum.
Constraints are not applied to the sweeps. The profiler shows the raw model surface across each factor's full range, so some points along a sweep may fall outside the constraints you defined. Read it as a view of the model, not as a set of runnable suggestions.
You can sweep up to 16 factors at once (it opens with a few by default). Add more from the X Axis selector.
Availability
The Prediction Profiler needs a trained model and a fresh prediction on the row you want to profile — the same conditions as the rest of the Predictions section. It is being rolled out gradually, so it may not yet be visible in your workspace.
Where to Go Next
Insights & Predictions — how to query a prediction, which the profiler builds on.
Analytics & Visualizations — the other charts available during a run.
Running an Optimization — back to the section overview.