TutorialsFilling the Three Lists: Six Questions That Make It Clear
Two questions each for requirement, design and acceptance. Tick the common options, add a few sentences, and AI has a brief it can read.
Six sentences are enough to specify one tool: who has the problem, what version one does, what form it takes, how the page should feel, what example you will test with, and what result that example should produce. This lesson fills those six into one Word checklist.
Three lists, six questions
| List | Question | What goes in |
|---|---|---|
| Requirement | 1 Who runs into trouble doing what | Main user (myself, colleagues, customers, students, family, other) + the task + the hardest step |
| Requirement | 2 Which single thing version one should do | One thing only, as specific as possible |
| Design | 3 What form it takes and how the result is used | Product form + device + how to start + how the result is used |
| Design | 4 How the page should feel | One of four styles, or reference images |
| Acceptance | 5 What small example I will test with | An input you can paste, or a clear sequence of actions |
| Acceptance | 6 What result that example should produce | A line-by-line checkable answer, not "works" or "correct" |
Requirement list: who has the problem, and what version one does
Question 1 uses this sentence:
Whenever [who] has to [do what], the hardest part is [which step].
Question 2 allows one thing only:
Version one just helps me [complete one task].
In the worked case: while organising event material lists, hunting for duplicates line by line is painful; version one only removes duplicates and returns a copyable list.
One thing only is the least intuitive rule in the whole course, and the most valuable. Every time you want to add a feature, go back to the previous lesson: remove it — can the user still finish the key task?
Design list: what form, and how it is used
Question 3 has four parts, all tickable.
Product form (single choice): web page in a browser, WeChat mini program, desktop tool, mobile app, chat assistant, or "let AI suggest".
Device (single choice): computer, phone, or both.
How you start (multi choice): type or paste text, upload files or images, choose from options, press a button, or let AI suggest.
How the result is used (multi choice): view on the page, copy out, download a file or image, keep it for next time, or let AI suggest.
Anything uncertain can be "let AI suggest" — it knows the core task and will pick the easiest-to-try option with the fewest dependencies, and explain why.
Question 4 is the visual feel (single choice): simple and AI-arranged, formal for office use, light and colourful, large text with obvious buttons, or reference images.
If you attach references, write one line per image saying what to borrow — colours, button shape, or result-area layout. A bare image makes AI copy the text and irrelevant elements along with it.
Acceptance list: write the standard before building
Questions 5 and 6 come as a pair: one input, one result.
I will enter or do: [one concrete piece of material or a clear sequence]
I expect to see: [a line-by-line answer]
The practice example from the handout takes six lines of input (line 4 is blank):
海报
签到表
海报
讲义
签到表The expected output is three lines:
海报
签到表
讲义The counts must match too: 5 valid items, 3 after cleaning, 2 duplicates removed. "Valid items" means entries left after trimming whitespace and ignoring blank lines; ignored blank lines do not count as removed duplicates.
One example proves one thing. Afterwards, let AI add the checks this build actually needs: core flow, empty input, the key boundary case, re-running with changed input, taking the result, and data behaviour after refresh.
How to fill the Word form
Save your own copy: use "Save As" with your tool name first. Never edit the original template.
Tick and fill: one tick per single-choice group, clear the old one when switching; multi-choice groups accept several ticks; grey areas take text. Open it in desktop Word with editing enabled so the boxes work.
Add references if needed: click the image or caption area, then the "+" at its lower right to add a group — one image with one caption each.
Do not force an answer you are unsure about. Write "to discuss", and settle it during the build.
Who does what
| You own | AI owns |
|---|---|
| Product goal and scope trade-offs | Layout and visual style |
| Choosing between key options | Interaction flow and feedback |
| Judging whether it is usable | Technical approach and code |
| Deciding whether to publish | Necessary testing and fixes |
A checklist only has to be good enough to start; details get filled in during the build. Every time you make an important choice, have AI write it back into the checklist — that is how the final result and the paper requirement stay in sync.
The point to catch: finish the acceptance list before building. Expected result first, tool second — otherwise a pretty interface convinces you that "it opens" means "it is right".
Finding the Problem: Start From One Concrete Annoyance
What natural-language programming actually is, how to source a topic from your own week, and how to decide how much version one should do.
Build and Verify: Ship the First Version with Baidu Dumate
Install the tool, hand over the checklist and prompt, decide the open questions one at a time, then run your own acceptance table against what it built.