What's New

The changes in each release. Each version has its own DOI, so you can cite the exact release you used. Installed copies update automatically through their store.

0.24.0 (15 September 2026)

DOI: 10.5281/zenodo.22764163

The catalog grows from 108 to 115 tools. Select Cases, Weight Cases, Split File, Recode, Transpose and both dataset merges are now commands in their own right. As a result, every change you make to the data in the menus is recorded, and a point-and-click session can be replayed from the original data to the same numbers. A merge records a fingerprint of the second file and checks it before anything changes.

Undo now treats each change to the dataset as one step. Deleting a column, recoding or editing cells is undone in full, and variables keep their identity, so columns never shift without notice.

Imported data is protected better. Integers too large to store exactly as numbers are kept as text, with a note. Values that can't be read as numbers are counted, marked in the Variable View and mentioned in any analysis that uses them. Date columns are checked on entry, stored in ISO format, and sort and filter in date order.

When input is rejected, the message says why. Unknown or misspelt arguments are refused by name everywhere, including in the AI assistant and MCP. Requests that are too large are shortened with a visible note.

Large text files now load without freezing the window. The app also works with screen readers: the data grid announces cells and selections, and every chart has a spoken summary and an accessible table of its data. This was tested by hand with Orca on Linux and Narrator on Windows.

0.23.0 (31 August 2026)

DOI: 10.5281/zenodo.22199867

The catalog stays at 108 tools. This release lets you check the results on your own installed copy.

Help → Verify this installation repeats the reference checks that are run before each release, on your computer, and writes a report. It also runs without a window, with --verify.

Every result now names its data. The footer of each output shows a fingerprint of the dataset, with any active filter, weight or split. File → Export Reproducibility Bundle saves the commands, the fingerprint and the version numbers in one file. ChakataStat can re-run the bundle against your data and confirm that the numbers match.

The methods are documented. Every procedure page in the guide has a "How this is computed" section, and 48 definitions that were previously implicit are now written down. Two defaults changed to the usual choice: Cox regression uses Efron's method for ties, and mixed models use REML. Saved scripts without those options now give slightly different numbers, and one option restores the old behaviour.

More cross-checks. Every value with a Python reference was recomputed in R (267 of 267, with each difference explained), and the full set was compared across Windows, macOS and Linux. The results are in the guide.

The file readers are more robust. A damaged or deliberately malformed SPSS, Stata, SAS, Excel, Parquet or Arrow file now produces an error message. The readers are fuzz-tested continuously and checked against 84 difficult edge cases.

0.22.0 (26 August 2026)

DOI: 10.5281/zenodo.22112257

The catalog stays at 108 tools. This release improves charts, export and numerical accuracy.

You can finish a chart without leaving the app. The chart editor adds axis titles, axis limits, reference lines, marker shapes and sizes, gridlines, figure height, a text-size slider and logarithmic axes. A log axis with a zero or negative value is refused with a message naming the value. Every edit is saved in the chart's command, so the chart looks the same when you run it again.

Charts can be exported for publication. Any chart saves as SVG or PDF, with labels kept as selectable text, and PNG export lets you choose the resolution.

Numerical accuracy improved. Normal and chi-square p-values used to become exactly zero beyond about eight standard deviations and are now accurate much further into the tail, checked against a 50-digit reference. Matrix inversion no longer depends on the scale of the variables, and the GLM families are fitted with an orthogonal decomposition.

Smaller changes: chart colours meet the 3:1 contrast guideline in both themes, long labels in the Variable View end with an ellipsis, scrolling dialogs fade at the edge, and charts redraw when you switch theme.

The AI assistant shows its replies as they are written and caches its instructions to reduce cost. You can also ask it to explain any result in your output. Both supported services were tested live.

Earlier releases

Releases before 0.12.0 are not listed here; their notes are in the project's release documents. Versions from 0.20.0 on have DOIs, listed on the How to cite page.