Scoreplane: Model Eval
Threshold & Calibration Lab
$0.99 · Designed for iPad. Not verified for macOS.
New in 2.0: compare models with a DeLong test, bootstrap confidence intervals, exact scikit-learn-matching AUC, and a threshold finder. One purchase, all on device.
Scoreplane is a private model-evaluation lab for data scientists, ML engineers, risk and fraud teams, researchers and students. Import a CSV of labels and model scores and see exactly how a binary classifier behaves, on your iPhone or iPad, with nothing uploaded anywhere.
NEW IN 2.0
- Exact metrics: ROC AUC and average precision use every distinct score and match scikit-learn, including ties and sample weights
- Compare models: put several score columns in one CSV and see their ROC curves and scorecards side by side
- DeLong test: a paired test of whether two models' ROC AUC really differ, with 95% intervals
- Bootstrap 95% confidence intervals for ROC AUC, average precision, precision, recall, F1 and Brier score (repeatable fixed seed)
- Threshold finder: pick the operating point for recall ≥ X, precision ≥ X, false-positive rate ≤ X, maximum F1, Youden J / KS, or lowest decision cost, then apply it with one tap
- Threshold sweep export: one CSV row per distinct score with confusion counts and metrics
- Gains and lift table by decile
- Library of saved evaluations with backup and restore files
- Raw scores such as logits or margins are supported for ranking metrics
- Redesigned interface with the system tab bar, dark mode, larger text support and iPad layouts
EVALUATE AN OPERATING POINT
Move the threshold and the weighted confusion matrix, precision, recall, specificity, F1, MCC, balanced accuracy, flagged rate and decision cost update together. A stacked score histogram shows where the threshold cuts.
CURVES
ROC, precision–recall, reliability (calibration) and cumulative-gains charts, with ROC AUC, average precision, KS, Brier score, log loss and expected calibration error.
GROUPS
Apply the same threshold to every value of a group column and compare base rate, ROC AUC, precision, recall, false-positive rate and flagged rate, with a warning for small groups.
IMPORT AND EXPORT
- CSV import: label column plus one or more score columns; optional id, group and weight. Comma, semicolon and tab files, quoted fields and Excel byte-order marks are handled, and skipped rows are reported
- CSV template to start from
- PDF report with ROC chart, threshold options, model comparison and groups
- Scored rows CSV with a flagged column, evaluation JSON, threshold sweep CSV
- Built-in metrics guide with the formula behind every number
A clearly labeled, app-generated sample evaluation (two models, four segments) lets you explore every screen before importing your own data.
EVERYTHING INCLUDED
One purchase, no subscription, no in-app purchases. No account, ads, analytics or tracking. Your data stays on the device unless you share an export.
Scoreplane measures how imported scores agree with imported labels. It does not certify fairness, safety, legality or production readiness; review sampling, label quality, drift and decision costs with the people accountable for the model.
Ratings & Reviews
- This app hasn’t received enough ratings or reviews to display an overview.
Scoreplane 2.0 is a major upgrade:
- Compare several models from one CSV, with a paired DeLong test for the AUC difference
- Bootstrap 95% confidence intervals for AUC, average precision, precision, recall, F1 and Brier
- Exact, scikit-learn-matching ROC AUC and average precision (no more threshold grid)
- Threshold finder for recall, precision, FPR, F1, Youden J or cost targets, plus threshold-sweep CSV export
- Gains and lift table, log loss and KS statistic
- Library of saved evaluations with backup and restore
- Raw (non-probability) scores supported for ranking metrics
- Semicolon/tab CSVs, byte-order marks and skipped-row reporting
- New interface: system tab bar, dark mode, Dynamic Type, better iPad layouts, metrics guide
- Removed the generated side workspace and the 340-image plate library (the app is about 200 MB smaller)
The developer, Robert Corbin, indicated that the app’s privacy practices may include handling of data as described below. For more information, see the developer’s privacy policy .
Data Not Collected
The developer does not collect any data from this app.
Accessibility
The developer has not yet indicated which accessibility features this app supports. Learn More
Information
- Seller
- Robert Corbin
- Size
- 3.4 MB
- Category
- Developer Tools
- Compatibility
Requires iOS 17.0 or later.
- iPhone
Requires iOS 17.0 or later. - iPad
Requires iPadOS 17.0 or later. - Mac
Requires macOS 14.0 or later and a Mac with Apple M1 chip or later. - Apple Vision
Requires visionOS 1.0 or later.
- iPhone
- Languages
- English
- Age Rating
4+
- 4+
- Copyright
- © 2026 Robert Corbin
