Tokeny task verification: setup, outputs and human review

What do these checks establish?

On November 5, 2026, the Tokeny team checked the tasks behind six tutorials using Tokeny 1.7.6 on Windows with the online gpt-6-sol model. Inputs were synthetic, screenshots came from the actual client, and generated files are available to download. Some tasks passed only after model corrections and human review. These are inspectable examples, not a measured success rate across models or tasks.

Current getting-started instructions have been updated for 1.7.8. The task environment and screenshots recorded here remain version 1.7.6. English tutorial prompts are adaptations; the actual screenshots and files are in Chinese.

Environment and method

Six checks and downloadable evidence

Excel: why recalculate the output?

The six synthetic orders calculate revenue as quantity × unit price. The first output had numeric-type, formula-reference and chart-position issues. After human feedback and model corrections, Microsoft Excel calculated the workbook independently.

These values apply only to the downloadable teaching sample. Recalculation after changing input checks that the workbook contains working calculations. Tokeny's spreadsheet preview shows cached values and collapses empty rows; inspect full styling and charts in Excel. See the CSV report tutorial.

This is the actual November 5, 2026 task screenshot. The download is the corrected, recalculated workbook saved after restoring the original input.

Word: check facts as well as layout

The meeting minutes and project report were opened, repaginated and rendered in Word, each as one page. Review covered names, dates, actions and open questions, preserving that the November 12 launch still required review and the budget remained unapproved. A confirmed action means an agreed assignment, not completed work.

The tasks encountered a mistyped path and an XML validation issue, then continued after corrections. Downloadable files contain the reviewed results; they do not demonstrate error-free first attempts. See meeting notes to Word.

Apply the same checks to your work

1. Preserve the original input and record the client version, model, required skills and MCP versions.
2. Specify the output path, format and observable acceptance criteria.
3. Check actual tool operations and saved paths, recording any corrections.
4. Open the output in its corresponding application and review facts, formulas or layout rather than relying on the completion message.
5. Keep the input, final file and review results together. Distinguish revised output from a successful first attempt.

These cases did not cover macOS, Linux, offline models, every provider or authenticated third-party account actions. The protocol initialization and tool-discovery checks for other MCP servers are also distinct from completed business tasks. See the product FAQ and current documentation for current workflows.