Upload a CSV with a true yes/no outcome and a predicted probability, and get the precision, recall, F1 and cost at every candidate cutoff — plus the optimum under each named criterion and the cost ratio each one quietly assumes. Free.
Free analyses run on up to 10,000 rows. Larger files are randomly sampled to that size — sign up to analyze your full dataset.
Sweeping every cutoff and costing the trade-offs...
Sent to — Metric curves across every cutoff, the optimum under each named criterion with its implied cost ratio, the confusion matrix at the chosen cutoff, a bootstrap optimism correction, R code, and AI insights.
Analyze another fileOne class is treated as positive, then every distinct value of the score is taken as a candidate cutoff: a single descending sort plus cumulative sums gives the true-positive, false-positive, true-negative and false-negative counts at every cutoff at once, and from those the precision, recall, F1, specificity, accuracy and — when costs are supplied — the expected cost per row. Named criteria are then located on that sweep: maximum F1, maximum Youden's J, minimum expected cost, and the cutoffs that meet a target precision or a target recall. For each criterion the analysis also computes the exact interval of cost ratios (cost of a false negative over cost of a false positive) for which that cutoff is the cheapest of all, which turns a criterion into a statement about beliefs. Because a cutoff selected on a sample and scored on that same sample is optimistic, an Efron optimism bootstrap over 200 resamples — each picking its own cutoff by the same rule and being scored back on the original rows — estimates the optimism and corrects the reported F1 and expected cost, and reports the 95% interval of where the cutoff itself landed. When the score is a probability, the cost-implied break-even cutoff is reported alongside the empirical minimum as a calibration check.
Use it when you have a real yes/no outcome and a model score or probability, and you need to decide where to draw the decision line — especially when a false alarm and a miss do not cost the same, or when the positive class is rare and the default 0.5 is clearly wrong.
Not for judging whether the score is any good in the first place (that is the ROC analysis, which is threshold-free), not for multi-class outcomes, and not a substitute for a genuine holdout: the bootstrap measures the optimism of in-sample cutoff selection but a cutoff validated on unseen rows is always stronger evidence.
Built for: Analysts, data scientists, risk and operations owners who have to turn a model score into an actual decision rule
Typical data source: A table with a true binary label and a predicted probability or score per row, plus a rough sense of what each kind of mistake costs
Any table with a true yes/no outcome and a numeric score per row. For example, a churn model's scored customers:
Minimum 20 rows · Best with 500-50,000 rows with at least 10 of each class; the cutoff is poorly determined below a few hundred rows and the report says so
Standard-library analysis: where should the cutoff go on your model score, given what YOUR errors actually cost? Map the true yes/no outcome and a predicted probability or score, and every candidate cutoff is swept — precision, recall, F1, specificity, accuracy and expected cost at each. The report gives the optimum under several named criteria (maximum F1, Youden's J, minimum expected cost from your own cost of a false positive and a false negative, and the cutoff that meets a target precision or recall), the metric-versus-threshold curves as the headline chart, and — for every cutoff — the cost ratio it implicitly assumes. If no costs are supplied, the criteria are shown side by side rather than one being picked for you.
Precision, recall, F1 and specificity traced across every candidate cutoff, with markers at the chosen cutoff and at the default 0.50 — the distance between them is the whole argument for computing a cutoff instead of accepting one.
Where each named criterion puts the cutoff and what it buys, plus the range of cost ratios that makes each one the cheapest choice — read that column as a belief you either hold or do not.
The four confusion counts and every rate at the leading cutoff, including how many rows it flags, with accuracy shown only against the majority-class baseline.
The bootstrap optimism correction, the interval the cutoff itself moved over, the class-balance verdict, and the calibration check — what the reported performance does and does not entitle you to claim.
Plain-English interpretation — what the numbers mean, what's significant, and what to do next.
Where should I set the cutoff given MY costs?
Map your true yes/no outcome and the model's predicted probability, and supply what a false positive and a false negative each cost you. The analysis sweeps every candidate cutoff and returns the one that minimises your expected cost — with the confusion matrix, precision and recall it buys, and how much cheaper it is than the default 0.50.
See our FAQ for details on pricing, data privacy, and how the analysis works. Every report includes a Methodology section showing the statistical test, assumptions checked, and diagnostics run.
Run any analysis on your own data — validated R analyses, interactive reports, AI insights, and PDF export.
Try Free — No Credit CardTell us what went wrong, in your own words. We capture the page you're on automatically, so no need to describe where you are.