Week 6 · Run Your Study

01 · Today

• • Warm-up Get the study right, fast A quick pass: measure before AND after, name a success line, drop the leading questions, add consent + Supabase. Done fast — so the rest of class is real data. First
  • Today 1 Play and review each other's pages Trios of three: play each other's pages for real, produce live pre/post data, and give each other one concrete fix. Together
  • Today 2 Let AI read your Supabase The heart of today. Point Codex / Cursor at your own table and let it do the stats and spot the pattern — you crunch nothing by hand. Core
  • Today 3 Interview a few people Numbers say what changed, never why. Sit down with 3–5 people and actually ask. Depth
  • Leave with Real responses, and an AI read of them By the end: pre/post data in your Supabase, you've watched AI analyse it, an interview plan ready, and the link out recruiting. Due today

02 · Get your study right (fast)

Read your hypothesis again. Almost every one of you wrote "after people use my page, they'll be more something" — more aware, more understood, more able to spot the bias. That word — more — is a change: it compares a later state to an earlier one. But your page measures people only once, after they've played, with nothing to compare against — and one after-reading can never show a change.

Nail this one sentence: to prove a "change", measure the same thing twice — once BEFORE they play, once AFTER. That's a pre/post survey: the same ~5 questions, tied straight to your hypothesis, asked once before and once right after; the difference is your result. Have AI design it — and keep both rounds identical, or the two readings can't be compared.

我在做一个关于【一句话写你的研究问题 / 社会议题】的网页,我的假设是【写你的假设,比如「玩过之后,人会更容易察觉到 XX」】。请帮我设计一套很短的小问卷,关键是要能测出「变化」:同样的问题,在用户玩我的网页之前问一次、玩之后再问一模一样的一次,这样我才能对比前后、算出到底变了多少。

要求:
1. 紧扣我的研究问题和假设出题,只问真正能反映「变化」的东西,别问没用的;
2. 题目少而准,5 题左右就好,别让人填到烦;
3. 前测和后测用完全相同的题目,只有题目一样,两次才能对比;
4. 尽量给我能直接量化的题型(比如 1–5 打分),方便之后算前后差。

先把题目列出来给我看,我确认没问题了,你再帮我做进网页。我零基础,讲清楚每步。

Two quick guardrails before you collect a single response. ① Name your success line now. Write down, as a number in your spec.md, what result would count as your hypothesis being true — e.g. "post average at least 1 point above pre". Draw the line before you see the data, or in July any result can be spun to fit your story. ② Don't lead your users. A question like "Don't you think this is unfair?" tells people the answer you want, so they just agree and the data is worthless — have AI rewrite any question whose answer is guessable into a neutral or open one.

03 · Consent + Supabase

You're about to collect data from real humans, so the first screen they see is a short consent screen: plain words, four points — it's anonymous, it's voluntary, they can quit anytime, and the data is only for this class research. They tap "agree" before anything else loads.

请帮我的网页加一个简短的「知情同意」第一屏:用户第一次打开时,先看到这一屏,读完点「同意」才进入正式内容。

这一屏用大白话说清楚四件事:
1. 匿名——我们不会收集你的姓名等能认出你是谁的信息;
2. 自愿——参不参加完全由你决定;
3. 随时可退出——中途不想填了,随时关掉离开就行;
4. 数据用途——你的回答只用于这个课堂研究,不会作别的用途。

请用纯前端(HTML/CSS/JS)实现:没点「同意」就不显示后面的内容。我零基础,改完告诉我加在了哪里、怎么自己测一遍。

Then wire both the pre and the post answers into the Supabase table you already set up on 6/21, so everyone's responses pool in the cloud and you watch them from one place. Two things make the data usable later: a timestamp on every row, and a field that marks each row as pre or post — without it you can't tell the two readings apart.

我之前已经把网页接到了 Supabase(有一张表在收数据)。现在请帮我把这套前测 / 后测问卷的提交也接到我已有的那张 Supabase 表里,让所有人填的答案都汇到云端、我在一个地方就能看到。

具体要求:
1. 用户提交前测和后测时,都把答案写进 Supabase;
2. 每条数据自动带一个时间戳(什么时候提交的);
3. 每条数据都要有一个字段标明这是「前测」还是「后测」,这样我之后才分得清、能对比;
4. 沿用我现在已有的那张表和 key,别新建一套。

我零基础,如果要改哪个文件、填哪个配置,带我一步步做;改完告诉我怎么自己测一条数据有没有真的进到 Supabase。

04 · Play and review each other's

You don't have to wait for outside users to get data — make a first batch right now, in the room. Split into groups of three: play each other's pages for real and give each other one concrete fix. By the time the trio is done, everyone has genuine pre/post rows in their Supabase and a real piece of feedback to act on.

• • Group Trios of three Split into groups of 3. You each play both teammates' pages — and they each play yours. Three
  • Play it Pre → play → post, for real Open a teammate's link and do the whole flow honestly: answer the pre, actually play, answer the post. No lazy click-through — the rows land live in their Supabase, and junk clicks make junk data. No skipping
  • Review it One concrete fix each After you play someone's page, give them ONE specific improvement — a confusing question, a flat moment, a leading item. Not "nice job" — one real thing they can change. Each person
  • Result Real rows + real feedback When the trio finishes, everyone has a handful of genuine pre/post rows to analyse today, plus two concrete suggestions to act on. Live data

05 · Let AI read your Supabase

This is the heart of today — and you do not crunch any numbers by hand. Point Codex or Cursor at your own Supabase, through the Supabase MCP you already set up, and let it read the table for you: how many people played, the pre vs post average and how far it moved, what people picked most, and whether the pattern your hypothesis predicted is actually showing up.

We'll do it together first. One of you connects their Supabase live, and the whole class watches the AI read the data we just collected in the trios and talk through what it finds. Then you run the same prompt on your own table.

请通过我配好的 Supabase MCP 读我的数据表,帮我做描述统计:一共多少人、前测和后测的平均分各是多少、变化多大、各个选项的分布;再对照我的研究假设【一句话写你的假设】说说初步看到了什么。

数据很少的话就如实说样本太小、别编结论。

06 · Interview a few

Numbers from a survey tell you what changed, never why. So pick 3–5 people and actually talk to them: "why did you choose that?", "what were you thinking while you played?", "did anything change your mind?" Open questions, a real conversation — not a form.

What you ask in the interview decides what you can write in your paper — and what evidence you'll have to back up your findings. So think the questions through before you sit down, not after. Have AI draft them from your research question.

请根据我的研究问题【一句话】,帮我设计 3–5 个开放式访谈问题,要能挖出「为什么、当时的感受、有没有改变想法」,别用一句是/否就能答完的问题。

07 · Leave with

The trio gave you a first taste of data. Two hard rules now carry the project home. The first is the one nobody wants to hear, and the one that saves your study.

The second rule: send the link out today and keep recruiting. The trios are your first rows; real outside people are next. Post your URL and ask them to take it — aim for a target of N ≈ 15–20 people who finish both the pre and the post.

• • Recruit Classmates & friends first Fastest source. Send it to people around you and ask them to do both halves — before AND after — not just click once. Start here
  • Recruit Moments & group chats Post the link with one line on what it's about and that it's a short anonymous class study. Make it easy to say yes. Wider
  • Recruit Xiaohongshu & relevant communities If your topic has a community that cares about it, share it there — those people give you the most honest, on-topic data. On topic
  • Target N ≈ 15–20 complete pairs A "complete" response is someone who did the pre AND the post. That pairing is what lets you measure the change — chase complete pairs, not raw clicks. Goal

Take it home. You leave today with the whole loop in place — pre/post survey, a success line in spec.md, consent + Supabase, and the AI read you just learned. After class, keep the data coming and run the analysis again as it grows.

• • Homework Keep sending the survey out Past the trio, recruit real outside people toward N ≈ 15–20 complete pairs — someone who did the pre AND the post. Due
  • Homework Re-run the AI read as data grows Each time more rows land, point Codex / Cursor back at your Supabase with today's prompt and watch the picture sharpen. Repeat
  • Coming up Pool it and compare Later we'll bring everyone's data together and compare before vs after across the whole class. Not today — just so you know it's next. Later

08 · Before you go — class feedback

Two minutes, every class. Tell us what landed and what didn't — it directly shapes next week's class.