From a calculation to work you can reopen and share.

1. Enter a calculation.

Type 9/2 in the calculator and press Enter. The supported exact result is 9/2, equivalent to 4.5. Use the in-app Function Reference or Examples for the accepted syntax of units, matrices, native algebra and graphs; inserting an example and calculating are separate actions.

For an exact algebra task, rational_simplify((x^2-1)/(x-1),x) gives x + 1 while retaining x − 1 ≠ 0. The original expression remains undefined at x = 1. Read the operation's input and domain limits. Graphs are sampled views, not exact curves.

2. Find the statistical task and review its inputs.

Open Find a tool with Ctrl+Shift+P and search for statistics, mean comparisons, ANOVA or regression. Choosing a result navigates to the actual tool; it does not select a method, insert data or run a calculation.

1. For descriptive exploration, review your raw or occurrence-frequency data and save the dataset in the workspace before analyzing it.
2. For a mean comparison, deliberately choose raw saved observations or Summary statistics. Supplied sample SD is not variance, SE or known population SD. Known-SD z has a separate sigma input.
3. Choose the study-design-appropriate method, direction, confidence level and alternative. Pooled independent t and ordinary ANOVA assume equal population variances; Welch uses a different calculation. Paired tests require genuine paired rows.
4. Review the prepared inputs and assumptions, then explicitly run. For an interval without a hypothesis test, choose Two-sided interval only. Test alternatives do not change the established two-sided confidence-interval convention.

Try the known-SD example or the three-summary-group ANOVA example. Summary input does not reconstruct missing observations or prove model assumptions. The app can refuse a result outside its supported numerical bounds; keep the inputs and inspect the reason.

3. Keep the accepted result, then save the file.

Draft fields, reviewed inputs and accepted results are different states. An edit does not make an earlier result current. Keeping a dataset or result in the workspace is not the same as writing a file: use File → Save or Save As to save the supported workspace to disk. Exporting a report does not mark it saved.

Use a synthetic practice file first. Close and reopen it to inspect the retained work. Stored statistics and symbolic results are historical snapshots, not live dependencies or newly verified calculations. Deliberately Review/Run when you need a fresh result; opening alone does not calculate.

4. Choose what to share.

For an accepted supported 2D or statistical plot, use its contextual Export figure action. Choose PDF for vector plot geometry or PNG for a raster image, set the size and inspect the preview before exporting. Vector output preserves sampled geometry; it does not make it mathematically exact. Existing 3D image export remains raster.

For several results, open the existing Analysis Report workflow, select the records deliberately and choose PDF for searchable text, or the available HTML/text output. Review the selected contents and page preview. Report figures may be raster; no PDF/A, tagged-PDF or universal editor compatibility is promised. SVG is unavailable.

A report is a presentation, not an editable SumWise workspace. Keep the .sumwise file as well if you want to return to the supported calculation state.

5. Keep originals when a file needs an upgrade.

Compatible content can use ordinary Save. New record types can require a newer reader; a genuinely incompatible upgrade of a named file requires Save As to a different destination. Do not overwrite the original if you need it in an older version. Old-only files retain their compatible encoding.

Corrections to mean inference and exact binomial behavior do not silently rewrite saved results. Explicit Review/Run uses current methods. A stored provenance label is historical information, not authentication. The Store release can read supported saved symbolic records without Python; it does not include the private Python calculation backend.

Know the limits before relying on a result.

Mathematics and Ask interpretation run locally; Microsoft Store handles installation, trial, purchase and license verification. This does not promise indefinitely offline authorization. SumWise is bounded software, not a universal CAS or a substitute for professional validation. Sampling, units, assumptions and stated refusal limits still matter.

Choose English, Spanish, French or German for the app interface; interface changes apply on the next launch. Some technical blocks remain in English. Numerical syntax uses a decimal point even in a localized interface. This English website guide is not a claim of fluent-human validation of every translation.

Current release notes · Support · Privacy