Changelog (release notes)

Version 0.6.1 — 20 Oct 2026

Fractional factorials for 9 to 15 factors: the catalogue of 2^k−p designs now goes up to 15 factors — with 128, 64, 32 or 16 runs for 9 to 11 factors and 32 or 16 runs for 12 to 15 factors. For 8 factors there is also the half fraction 2⁸⁻¹ (resolution VIII).

Resolution IV fractional factorials are now analysed by default with one interaction per alias chain instead of main effects only, as in Design-Expert; a note at the model selection explains the reduction. Analyses of such designs with the default model change as a result.
The project explorer no longer scrolls sideways: long titles end in “…” and show the full name as a tooltip.
Labels in the half-normal and Pareto charts no longer overlap and are no longer cut off at the edge.

Design-Expert import: 2^k−p fractions are recognised as fractional factorials with their resolution, alias structure and matching model instead of as full factorials; specification limits are taken over correctly and no longer derived from the desirability. Please recalculate imported projects.
After the specification limits are changed, simulation and optimisation are marked as out of date instead of still showing results for the old limits.
No more crash when a mixture component is moved after switching projects.

Network licences (concurrent use): A licence can be activated on any number of computers; as many users can work at the same time as seats have been purchased. When all seats are taken, DoEStat starts as a viewer and checks again periodically. A seat is returned when the program closes and frees itself 30 minutes after a crash.

The strategy comparison states, for every design family, which model the efficiency and quality of the candidates were assessed against.
Numbers in all charts follow the interface language: German with a decimal comma, English with a decimal point (previously, e.g., “0.5” could appear on an axis next to “0,5” in the table). The values themselves do not change.

Space-filling designs for computer experiments: Latin hypercube (also under constraints), Sobol (Owen-scrambled), orthogonal-array Latin hypercube, sphere packing and MaxPro, also with categorical and discrete factors; existing designs can be augmented with space-filling points.
Supersaturated designs (Bayesian D/I) with an adjustable prior; they are analysed as far as the data allow instead of being rejected.
New Taguchi array L25 (5⁶); mixture designs with more than three components in contour and 3-D plots.

Kriging (Gaussian process) as an analysis method: selectable kernel, trend and nugget, diagnostics, categorical factors, suggestion for the next run. Optimisation, simulation and intervals are computed on the Gaussian process.
Pass/fail and class responses: classification page with ROC curve, threshold and metrics; Firth correction plus probit and cloglog links; multinomial and ordinal logistic regression; text responses. Count and proportion responses are computed in the appropriate distribution model.
Model selection: ridge regression, SVEM ensemble, bootstrap stability check, quadratic mixture terms, cubic mixture models, AICc/BIC as in Design-Expert; blocks as a random effect with variance components.
Tolerance interval alongside confidence and prediction intervals; further transformations and a freely selectable λ.

Desirability rebuilt: one shared goal system for responses and factors, explorable interactively and locally optimisable; goal types “in range” and “none”; Pareto front as a chart.
“Account for scrap”: the optimisation takes the share outside the specification limits into account, so the optimum moves away from the limit towards the inside (on by default, can be switched off).

Simulator with further distributions and measurement scatter; excluded regions (constraints) can be drawn graphically.
Design-Expert import now also takes over constraints, split-plot structure and the model and transformation of each response.
Report in three depths (short, medium, detailed) as Word, PDF, PowerPoint, HTML or Markdown; every table can be saved as CSV.
New example projects, English manual, manual in the browser, Python integration with its own manual chapter.

Online licensing: enter the licence key under Help → Licence and activate it online; “Release on this computer” returns the seat. The licence is checked at start-up and every 24 hours after that, and keeps working without internet for up to 14 days. The licence file remains available for computers without internet access.
Trial: a 14-day trial key replaces the previous local 30-day trial. You can request a trial via doestat.com. Without a licence, DoEStat starts as a free viewer.
Automatic updates: an installation activated online checks for new versions at most once a day and asks before installing.
Updated licence agreement (EULA) and terms and conditions: you will be asked to accept them again at the next start.

The GxP package is hidden by default and can be set for all users via a machine policy.
Electronic signature (21 CFR Part 11): Help → Sign project… with meaning, comment and password; signature status in the header and the report; a reason is required for the first change to a signed project; new checks in the OQ self-test.
The audit trail records the data source and model for each response and the old and new values of changed measurements and settings; bulk exclusions appear as a single entry; two gaps in the verification closed. New manual section on the audit trail and the duty to give reasons.

Charts are larger, resizable and only render once they become visible; new chart inspector; overlay contour plot with draggable contour lines; point colour by factor level.
The project name is always shown at the top and can be renamed with a click; projects open by double-click and can be reopened after a crash.
Several responses are computed side by side considerably faster; cancelling also works in the middle of a calculation.
Labels and numbers follow the interface language; the English interface no longer contains leftover German text.
Stability and security: export protects against formula injection in Excel, damaged or foreign import files are rejected, report export and page changes no longer block; optional crash reports only with your consent.

Calculation errors: very small p-values became 0; regression with extremely poor conditioning; wrong degrees of freedom for blocks; predicted R² of 1.0000 and G-efficiency above 100 % for saturated models; rank-deficient designs and models silently returned wrong values; the optimisation did not respect specification limits; the transformation and two analysis settings were lost on saving. Please recalculate existing analyses with 0.5.1.
Import errors: an empty row in Excel shifted all following runs; “Infinity” was read as a measured value; UTF-16 CSV produced one run too many; typed decimal numbers were read as thousands separators; mixtures from .dxp files with a total other than 1 were shown scaled down.
Copying and pasting in the data table changed measured values.