Interactive measurement explorer
Drag to zoom · hover to inspect · double-click to reset · use the toolbar to pan or save PNG.
Select a model
| Parameter | Estimate | ± 1σ | 95% interval |
|---|
Residuals
Measured Y − fitted Y. Look for trends, unequal spread and outlying points.
Model comparison · first series
Ranked by AICc using the same rows and likelihood. Lower is better. Failed or unconverged fits are excluded from ranking. Select a row to inspect its fit. High-degree fits can overfit; validate on independent data.
| Model | R² | RMSE | AICc | Status |
|---|
Detailed & robust statistics
Statistics cover the included measurements in every selected series, independent of model fit success. Tukey fences flag potential outliers; no data is deleted.
Correlation & inference
Pearson p-values assume independent bivariate-normal observations. Spearman p-values use a t approximation with average ranks for ties. P-values are unadjusted for multiple comparisons; sample-mean intervals do not represent instrument uncertainty.
Fit diagnostics
Durbin–Watson is calculated in ascending X order and is descriptive here. Jarque–Bera uses an asymptotic χ² approximation for n ≥ 20; its p-value is a residual normality diagnostic. Nonlinear parameter intervals are local approximations and may be unreliable near bounds.
Y distribution
Pearson correlation matrix
Pairwise complete observations within the X range; correlation is not causation.
Imported measurement table
The worksheet displays every imported row in pages of 100. Search filters the worksheet view only; analysis uses your X/Y selections and X range. Blank, missing and nonnumeric X/Y rows are counted and excluded for each series. Outliers are retained.
Prepare a derived column
Original columns remain unchanged. Missing and out-of-domain cells stay blank in the derived column. Select the new column explicitly for analysis.
Graph studio
Uses selected X/Y columns and the X range. Plot without fitting a model.
Hypothesis testing
ANOVA uses all selected Y columns as independent groups. Paired tests retain complete row pairs. Independent tests retain each sample's numeric cells separately. Normal KS requires μ and σ specified independently of these observations. Chi-square uses paired count columns with matching totals and expected counts ≥ 5.
Multivariable regression & principal components
Select 1–12 predictors for regression or 2–12 variables for PCA. Uses complete rows across selected variables within the X range. Regression includes an intercept and requires a full-rank design. PCA scales each variable to unit sample variance.
Signal processing & peak analysis
Processes the first selected Y series in ascending X order; duplicate X values are averaged and reported. FFT requires uniform spacing and uses mean removal, a Hann window, radix-2 FFT with zero-padding, and coherent-gain amplitude normalization (frequency units are inverse X units). Autocorrelation requires uniform spacing. Baseline uses λ = 100,000, p = 0.01. Smoothing and derivatives use a cubic local polynomial.
Advanced results are exported separately from curve-fit reports. Changing the input data or selections clears advanced results.
Your analysis workflow
- Import. Choose a data file or a synthetic example. Select a worksheet for an Excel workbook. Adjust header, skip and delimiter settings for instrument exports.
- Select. Choose X, one or more Y columns, and optionally a Y standard-uncertainty column. Add units to axis labels and restrict the X range if needed.
- Fit. Choose a scientific model or enter your equation. For difficult nonlinear problems, enter starting values and bounds. Weighted fits use 1/σ² weights.
- Inspect. Read residuals, convergence warnings and parameter uncertainties. Compare models when you do not yet have a justified model.
- Export. Download results CSV, Excel, JSON, an HTML report, or a PNG/SVG figure. Data is never uploaded.
Supported data formats
Spreadsheets: .xlsx, .xls, .xlsm, .xlsb, .ods, .fods, SpreadsheetML .xml and HTML tables. “.xsls” is accepted as a misspelled Excel filename. A file must still contain valid spreadsheet data.
Text / Notepad: .csv, .tsv, .txt, .dat, .asc, .log and .prn with comma, semicolon, tab or whitespace-separated tables. Lines beginning with # or % are ignored. Decimal points are required.
JSON: arrays of records, arrays of rows, equal-length column arrays, or a data array inside an object. JSONL / NDJSON accepts one record per line.
Excel formulas are not recalculated. Cached formula values are used; formula cells without numeric cached values are excluded. Password-encrypted files, binary scientific formats (HDF5, NetCDF, MAT, Parquet), images, scanned PDFs and narrative Word documents are not supported in this release.
Advanced workspace
Worksheet: view, search and page through uploaded data immediately. Graph studio: independent plots, box/violin distributions, correlation heatmaps and 3D scatter. Hypothesis tests: two-sided t tests with confidence intervals, classical one-way ANOVA, tie-corrected approximate Mann–Whitney, normal KS with known parameters, and chi-square goodness of fit. Multivariate: ordinary multiple regression and standardized PCA. Signal: local polynomial processing, integration, FFT, autocorrelation, baseline correction and peak detection.
This release implements a selected Origin-style workflow, not the full Origin/OriginPro product. Native .opj/.opju projects, LabTalk/Origin C, image processing, surface fitting, general DOE, repeated-measures ANOVA, and proprietary instrument formats are not supported. Use SpectraLab for simultaneous spectral peak fitting and calibration. Statistical assumptions, sampling design and instrument calibration must be assessed by the researcher.
What the numbers mean
R² = 1 − RSS/TSS; RMSE = √(RSS/n). R² is undefined when Y is constant. Ordinary-fit covariance uses RSS/(n−p); weighted covariance treats supplied σ as known absolute errors. Nonlinear errors use a local Jacobian approximation and may be unreliable near bounds or with correlated parameters. They are not 95% confidence intervals.
AICc compares the same observations and error model, counting estimated residual variance for ordinary fits. AICc is unavailable with too few observations. Convergence does not prove a global optimum. The model library is broad but cannot cover every possible scientific equation; custom expressions extend it.
Privacy & deployment
No paid API, account or server is required. Dependencies are included locally. Reloading clears the session. Host this folder under /jakobura/ and link to https://www.afriremotelabs.com/jakobura/. Existing site access rules remain the responsibility of the host.
03 Export your research
Measurements, model parameters, diagnostics and publication-ready figures.