Inconsistent ROC confidence intervals: rocmetrics plot vs. manually extracted intervals

I am currently working with the rocmetrics class and its bootstrap-based confidence intervals. While the built-in plotting functionality works well, I observe a discrepancy when I try to extract and visualise the confidence intervals manually from the Metrics property.

Specifically, the confidence intervals shown by

plot(RO,"ShowConfidenceIntervals",true);

in the plot in red do not match the intervals I obtain when directly plotting the corresponding columns from RO.Metrics in blue, but the ROC itself matches.

The manually reconstructed intervals appear:

• narrower, and
• slightly shifted along the x-axis

compared to the intervals shown by the built-in plot.

Minimal Working Example

close all
rng(1)
nData = 500;
scores = rand(nData,1);
class = (scores+(rand(nData,1)*2-1)*0.4)>0.5;

RO = rocmetrics(class,scores,1,NumBootstraps=100);


plotObj = plot(RO.Metrics.FalsePositiveRate(:,1),RO.Metrics.TruePositiveRate(:,1));
hold on
fill([RO.Metrics.FalsePositiveRate(:,2);flipud(RO.Metrics.FalsePositiveRate(:,3))],...
    [RO.Metrics.TruePositiveRate(:,2);flipud(RO.Metrics.TruePositiveRate(:,3))],...
    plotObj.Color,'FaceAlpha',0.3,'EdgeAlpha',0);


plot(RO,"ShowConfidenceIntervals",true);

Observations

• The central ROC curve ((:,1)) aligns well between both approaches.
• However, the confidence intervals differ noticeably:
• The manually plotted intervals are smaller.
• They also appear misaligned in FPR (x-direction).
• There seems to be no direct documentation on the Metrics data using bootstrapping

Question

I would like to understand:

1. Why do the values in RO.Metrics.FalsePositiveRate(:,2:3) and TruePositiveRate(:,2:3) not reproduce the plotted intervals?
2. Are these quantities defined on a different grid than the plotted ROC?
3. Does the plot function perform additional processing (e.g., interpolation onto a common FPR grid)?
4. What is the correct way to extract and reproduce the confidence intervals programmatically?

Environment

MATLAB Version: 25.2.0.3150157 (R2025b) Update 4

Operating System: Windows 11

Statistics and Machine Learning Toolbox: Version 25.2

Additional Notes

My assumption is that the discrepancy might stem from how bootstrap ROC curves are aligned (e.g. interpolation to a common FPR grid before computing percentiles), whereas the raw Metrics values may correspond to threshold-based sampling. However, I have not found clear documentation on this.