Prepare a small bank with answers you can check
Before moving an employee question bank, run a rehearsal whose answers are visible and whose people are fictional. This pilot uses a made-up studio's document-release rules. It tests the software workflow rather than job competence or a real safety standard.
Download the files below into one directory. They form an independent English sample; the original Chinese demonstration files remain unchanged.
| File | Purpose |
|---|---|
| Rehearsal rules | The complete fictional rule sheet used to answer the questions |
| Twelve-question English DOCX | Normal Web upload and review |
| Source questions and answers | Neutral source for checking content or preparing another platform's format |
| Expected answers | Question IDs, types, correct options and boundary expectations |
| Two boundary questions | Separate missing-answer and option-order checks |
| Prepared candidates | Authorized MCP/CLI structured import, not the DOCX upload form |
| Two fictional participants | Ordinary member rows without credentials or contact details |
| File manifest | File sizes and SHA-256 digests |
| Sample notes | Format, identity and verification boundaries |
The normal bank contains eight single-choice questions and four multiple-choice questions. Four single-choice items present statements with English True and False options. They deliberately use the single-choice type; this pilot does not verify the separate judgment question type. Check any additional types your organization needs with separate samples.
Keep the supplied import syntax
Read the rule sheet first. The current DOCX parser requires literal 答案： and 解析： markers for the answer and explanation. The supplied file already includes them; keep the markers while editing the English content. Option keys and the answer key must continue to refer to the same choices.
The expected answer sequence for G001–G012 is:
B, B, B, C, AC, AC, AC, AC, A, B, B, A
These answers follow the fictional rule sheet. The shipped files were compared with the existing DOCX parser for question count, types, complete stems, options, answers and explanations. That file-level check does not prove a completed Web upload, an end-to-end assessment or another vendor's import. If you edit the source, review the answer key and regenerate the relevant manifest rather than reusing old hashes.
To inspect an individual download on macOS, for example:
shasum -a 256 training-standard.en.docx
Compare the digest with that filename's manifest entry. A matching digest establishes file identity, not answer correctness. The sample contains no passwords, activation codes or actual employee data.
Publish the normal bank and check personal practice
In a personal space, create a rehearsal bank and upload the normal DOCX. Open each candidate and compare it with the source. Check that multi-answer items retain both correct choices and that the False statements have not changed meaning. Repair or exclude uncertain candidates before accepting them.
Publish the import, then publish the bank version. Confirm that the resulting bank contains twelve usable questions. A candidate count and a published version are different results; record both during the rehearsal.
Start a short practice session. Deliberately give one wrong answer to a fictional question, inspect the feedback and explanation, then find the question in mistake practice. Answer it correctly on a later attempt and check the recorded outcome. Retain the real result you observe rather than assuming that publication proves the learning workflow.
Review the two boundary cases separately
G013 intentionally lacks an answer. It should remain blocked until a human has explicitly supplied and reviewed one. Do not let an assistant silently fill the omission just because the source rules make an answer seem likely.
G014 has an option that says “Both A and C.” Its meaning depends on those positions. Inspect it manually, set fixed option order and verify the displayed result before any publication. English positional phrases may not trigger the current parser's automatic warning. An empty issue list is not permission to shuffle the choices.
Keep the boundary import separate from the normal twelve-question bank. The files are designed to expose review decisions, not to demonstrate that all source errors can be repaired automatically.
Arrange one organization rehearsal with two participants
Use an organization you are authorized to operate with sufficient active service. The owner and two ordinary participants occupy three member places; existing active or pending members also count. Prepare files and access before activating a short trial. Check the actual capacity, duration and payment conditions on the pricing page.
For Web-only use, add the two fictional participants through the normal member interface. For authorized MCP administration, use the participant JSON in the AI walkthrough. A member row does not create an owner or administrator. Participants or their administrator handle normal password setup privately; credentials do not belong in the files or an AI chat.
Publish a bank version in that organization and verify the actual participant roster. Create a future exam window, a per-person duration and a clear answer-release policy. For this rehearsal, twelve questions at ten points each give a total of 120; a chosen passing score of 72 is only a workflow-test setting.
Read the timezone shown by the form and verify the intended instant with both participants. English interface text does not change the deadline. Have each person answer and submit under their own identity. This guide does not authorize an AI assistant to answer for employees.
Inspect a retake and reconcile the result file
Use the supported attempt or retake settings for your rehearsal and verify the score-selection policy. Record the original attempt, the later attempt and which one the report uses. Check absent and unsubmitted participants as well as successful submissions. Extending the overall availability window should not be assumed to extend an already-started person's deadline.
Read the results and export the available JSON or CSV. Match identifiers, submissions, scores and the configured assessment to actual participant actions. An exported report and a restorable system backup serve different purposes; the data portability guide explains supported restoration and historical archives.
Finish by narrowing or revoking any assistant authorization you added. Keep the rehearsal records you need and manage the test bank through normal product controls. Do not delete or change unrelated banks, members or exams.
For a product shortlist, use the small-team guide. To compare this work across vendors, adapt the neutral source into their documented formats and use the workflow checklist. The Learnweft DOCX and prepared JSON are not universal import contracts.