ACTUAL VS PREDICTED

Model Evaluation Tools

Compare predictions with observed values using explicit metrics and row-level errors. These calculators evaluate values you provide; they do not fit a model or make a quality judgment about a model’s intended use.

AVAILABLE CALCULATORS

Choose the metric you need

Use the same observations for actual and predicted values. Choose columns directly and inspect validation errors before calculating.

01RMSE calculatorRoot mean squared error, MSE, MAE and per-row errorsAvailable 02R-squared calculatorR² from observed and predicted values, with SSE and SSTAvailable 03Compare regression modelsRMSE, MSE, MAE and R² on the same observed casesAvailable 04Residual plot generatorResidual versus supplied predicted values with table exportsAvailable

STRUCTURED OUTPUT

Validate JSON model output

Check a local JSON response against a Draft-07 schema before reviewing it alongside other model-quality measures.

05LLM output validatorValidate JSON documents and JSON Lines against a JSON SchemaAvailable

RUN COMPARISON

Compare saved evaluation runs

Compare baseline and candidate LLM evaluation runs by stable case IDs. Review added, removed and matched cases, regression rate and pass-rate delta without calling any model.

06LLM regression comparisonPass/fail/skipped/error/unknown status transitions and comparability gateAvailable

Interpret metrics in context

RMSE and MAE use the same units as the observed values. RMSE penalises large errors more strongly because errors are squared before averaging; MAE averages absolute error. R² compares residual variation with variation around the observed mean. Compare models on the same observations and outcome scale, and never use these metrics alone to claim future performance. Compare multiple models

Calculate RMSE, MSE and MAE · Calculate R-squared