Week 7 · Write Your Paper Now

Let's be honest before we start. Across all eight of your projects, the number of real participants right now is basically zero. So nobody in this room can write a real Results section this week — and that's nothing to hide. It's just where everyone's starting from, and it decides exactly what today is for.

So here's the single job, for everyone, no exceptions: write an Expected-Results placeholder abstract and lock your Methods. Results stays empty — one honest line describing the table you'll fill later. Making up a number to fill it is the one move that can sink the whole paper. We'll say that more than once today.

And the textbook today isn't some model paper — it's your own eight projects. We'll sort them onto three starting lines, take three real abstracts apart move by move, and turn the Methods you can already write into the thing that unlocks the rest. Every grey box is a Chinese prompt you can copy into Codex or Cursor. Keep this page open. Stuck? Mia is here.

01 · Today: the honest truth

One idea runs through the whole class: real data is basically zero across all eight projects, so nobody writes a real Results section today. That's just where everyone's starting from, not a failure. Today you do one thing: write a placeholder abstract and lock your Methods. Here's how today goes:

No real data yet? Good — that's literally everyone. Now do not invent a single number. Writing a "reasonable-looking" result before the study runs is fabrication, and it can fail the entire paper. Today Results is one honest placeholder line, nothing more. The moment real responses land, you come back and fill it — that's next class, not now.

02 · Which starting line are you on?

Eight projects, but only three real situations. Before you write a word, find which starting line you're on — it decides your very first move today, and everyone's first move is different.

Three starting lines — find yours: A · Main line undecided can't lock the contribution sentence yet first move -> make ONE choice on paper, then write B · Tool ready, no people the build works; the sample is empty first move -> go recruit — even 5 people to test the flow C · Design still leaks a confound would spoil the data first move -> fix the design on paper and freeze it

Your first move, by line: A → answer the multiple-choice on paper (measurement vs. intervention, which unit, which scope); B → walk out and recruit — even five people to run the flow end to end; C → redraw the design and freeze it before a single response comes in. Claim your line now — everything else today follows from it.

03 · Paper 101: the parts and the words

Before we touch your own projects, the basics — no project here, just the map. Your paper is only a few blocks, always in the same order: that's IMRaD. It runs on a handful of words that show up in every paper. And it opens with an abstract. Get these three things down first and everything else today has somewhere to land.

First, IMRaD — the four parts, and what each is for. Every paper is these four jobs, in this order. Here's one plain definition each — plus what you can already lift straight from your own midterm draft.

IMRaD — the four blocks every paper has, in this order: I · Introduction the social problem + why it matters + your RQ M · Method exactly what you did — so someone could redo it R · Results what happened — facts only, no interpreting D · Discussion what it means + the limits + the so-what (the Literature Review lives inside the Introduction)

Next, the words every paper uses. You'll hit these in every study you read and write. Most of you haven't seen them before, so here they are in plain terms — English and 中文.

Must-know terms · 英文 / 中文 / what it means: Research Question (RQ) 研究问题 the one question your study answers Hypothesis 假设 your prediction of what you'll find Independent Variable (IV) 自变量 the thing YOU change or control Dependent Variable (DV) 因变量 the thing you MEASURE for change Pre-test / Post-test 前测 / 后测 the same measure, before and after Sample / N 样本 / 样本量 who took part / how many people Descriptive statistics 描述统计 means, percentages — plain summaries Literature Review 文献综述 the prior studies you build on Research Gap 研究空白 what prior work still hasn't done

Last, the abstract — what it actually is. The abstract is your whole paper squeezed into one paragraph (about 150–250 words): one quick pass through background → question → method → finding. Here's the twist: you write it last, after everything else is done, because you can't sum up a paper that doesn't exist yet. So how do you draft one today, before your data is in? That's the next section — five moves, fill in the blanks.

04 · The abstract in 5 moves

The abstract is your whole paper in one paragraph — and it's a fill-in-the-blank, not a blank page. Five moves, in order. Everyone writes the same five sentences; only the content changes.

The abstract, in five moves — fill each blank: 1 · Hook the real-world pain + the common misread of it 2 · Gap what existing studies / tools still don't do 3 · Artifact what you built, in one sentence 4 · Method how you measure it: pre/post + N + the outcome 5 · Expected what you expect — or [pending data] hard rule: move 5 is "we expect..." or [pending data] — NO ONE writes a result number today

Moves ① and ③ you can write straight from what you already know. Move ④ you can write in full, because your tool is already built — the design is real, not a guess. And move ⑤ is where papers get faked, so guard it hardest.

Move ⑤ must never contain a result number — and AI will slip one in if you let it. The last sentence states what you expect and why, then ends in "we expect…" or [pending data]. Any digit that looks like a finding — a mean, a percentage, a p-value — is fabrication before the study has run. Read move ⑤ back to yourself: if there's a number in it, delete the number.

05 · Every project's abstract (EN ⇄ 中文)

Here are all eight abstracts from our class. Find yours. Use the toggle to read it in English or 中文.

Clean-Heating Transition Sandbox · 清洁取暖转型沙盘Cluster A
In Baoding's coal-to-clean villages, a farmer can afford to install a heat pump but not to run it once subsidies taper, so going back to scattered coal reads as backwardness, not budget arithmetic. Prior work measures this squeeze at the policy level but rarely lets one person feel a single winter's math. We built the Clean-Heating Transition Sandbox, a browser game where you run one farm household across five winters, choosing coal, gas, electric, a heat pump, or insulation, with surplus, energy burden, and emissions updating live. Runs end in transition, bankruptcy, or a fine; a pre/post design [N = 15-20] captures players' heating-cost estimates, their sense of affordability, and whether they blame a return to coal on structure or character. We expect cost estimates to rise and blame to shift from personal failing toward budget constraint [results pending data].
在保定的煤改村里,农户改得起一台热泵,却在补贴退坡后烧不起——于是有人悄悄烧回散煤,这常被读成观念落后,而不是账本上的预算理性。既有研究多停在政策与区域层面,很少让一个人真切算一遍自家一冬的取暖账。我们做了「清洁取暖转型沙盘」:一款浏览器游戏,让你扮演一户华北农家,在五个冬天里于散煤、天然气、电、热泵、保温之间做选择,年盈余、能耗负担率与碳排放实时跳动。每局以转型成功、破产或被罚收场;采用前后测设计([N = 15-20]),测量玩家对取暖花费的估计、对可负担与公平的感受,以及把返煤归因于结构还是个人品性。我们预期,玩过之后取暖花费的估计上调,返煤的归因从「个人不争气」转向「预算约束」[结果待数据]。
Mayor Climate Game · 市长气候游戏Cluster A
Cities face rising heat and cooling demand, forcing trade-offs between cheap high-carbon power and cleaner but costlier options—trade-offs that stay invisible at the moment of choice. Professional simulators (WRF/CFD/GIS) are too high-barrier for a classroom, while public climate games rarely record what players actually decide. We built the Mayor Climate Game, a browser serious game in which players spend a fixed budget placing coal, wind, and ground-source-heat facilities on a city grid while average temperature, wind speed, and carbon emissions update in real time. Framed as an intervention and using a pre/post design [N = 15-20], we test whether real-time micro-climate and carbon feedback lowers undergraduates' preference for cheap high-carbon coal and improves their reasoning about the cost–carbon–ventilation trade-off. We expect fewer coal placements and higher trade-off awareness after play [results pending data].
城市不断升温、制冷需求节节攀升,迫使人们在廉价的高碳电力与更清洁却更贵的选项之间权衡——而这种权衡在做选择的那一刻往往是看不见的。专业模拟器(WRF/CFD/GIS)对课堂而言门槛过高,面向公众的气候游戏又极少记录玩家究竟做了什么决定。我们做了《市长气候游戏》:一款浏览器端的严肃游戏,玩家用一笔固定预算在城市网格上布置煤电、风电与地源热泵设施,平均气温、风速与碳排放随之实时刷新。作为一项干预、采用前后测设计 [N = 15-20],我们检验实时的微气候与碳反馈能否降低本科生对廉价高碳煤电的偏好,并改善他们对「成本—碳排—通风」三者权衡的推理。我们预期,游玩之后煤电布置减少、权衡意识提升 [results pending data]。

'Framed as an intervention' is what picking a main line buys you.

Anonymous Self-Seeking Support Space · 自我认同匿名支持空间Cluster A
Teenagers questioning their sexual orientation often carry the confusion alone, not ready to say it out loud to a counselor, parent, or friend. The stress gets read as being about the identity itself, when research points instead to stigma and the absence of low-pressure places to ask. Online spaces can help, but rarely hold anonymity and safety together. We built a single-page web tool: a student picks one worry—"I'm not sure who I like," "I'm scared people will find out"—reads a non-judgmental reply and a safer next step, runs a short self-check, then can ask a constrained AI that only gives general support and never diagnoses or labels. With [N = 15-20] high-school users, we measure whether they feel understood and can name a safer next step. We expect a shift toward feeling understood [results pending data].
很多在探索自己性取向的高中生,把这份困惑一个人扛着——还没准备好当着辅导员、父母或朋友的面说出口。人们常把这种压力归因于"性取向本身",但研究指向的是污名,以及缺少一个低压力、能随口问一句的地方。线上空间能帮上忙,却很少能同时守住匿名与安全。我们做了一个单页网页:学生选一个当下的困惑(比如"我不确定自己喜欢谁""我害怕别人知道"),读到一段不评判的回应和一个更稳的下一步,做一次简短自评,再向一个受约束的 AI 提问——它只给一般性的支持和安全提醒,绝不诊断、也不给人贴标签。面向 [N = 15-20] 名高中生用户,我们测量他们是否感到被理解、能否说出一个更安全的下一步。我们预期被理解感会上升 [数据待补]。
Sandwich-Generation Time-Poverty Game · 夹心层时间贫困游戏Cluster B
Caregivers who simultaneously raise children and support aging parents tend to lose their own time first, yet this is usually framed as a personal time-management failure rather than a structural condition. Prior work documents the resulting stress but rarely turns the mechanism into something a person can feel. We built a single-page browser game in which a never-ending task list and an AI "priority system" repeatedly push the player's attempt to rest to the very end of the queue. Using a pre/post design, [N = 15-20] players answered the same five-item awareness scale before and after play, and we compare the mean change with particular attention to the "structural vs. individual cause" item. We expect awareness to shift from "personal failure" toward "structural condition," with the largest gain on the structural-cause item [results pending data].
同时抚养孩子、赡养老人的照护者,最先牺牲的往往是自己的时间;而这通常被归为个人时间管理不善,而非一种结构性处境。已有研究记录了由此产生的压力,却很少把这套机制变成一个人能亲身感受到的东西。我们做了一个单页网页游戏:一份永远清不完的任务清单和一套 AI「优先级系统」,把玩家想要休息的每一次尝试,一次次挤到任务队列的最末尾。采用前后测设计,[N = 15-20] 名玩家在游戏前后填写同一份五题觉察量表,我们比较其均值变化,并特别关注「结构性成因 vs. 个人成因」这一题。我们预期,觉察会从「个人失败」移向「结构性处境」,且在结构成因题上的提升最大 [results pending data]。

Move ④ is already writable — only ⑤ waits on data.

Reframe Destiny · 命运重构(命理性别叙事重构)Cluster B
Young people read a fortune-teller's script as a neutral map of their fate, quietly absorbing its gender roles: the same chart makes a man an "ambitious leader" and a woman "too strong" or "fated to harm her husband." Bias-detection tools target workplace language; fortune-telling apps optimize insight, not critique—neither hands a young reader the move to talk back. We built Reframe Destiny, a short bilingual web journey: pick a BaZi or astrology chart, read one traditional gendered reading, and tag the bias you see—marriage centrism, husband-harming, fear. The reader then compares an AI-reframed version—a constrained teaching script, not an oracle—and writes one line in a "Court of Destiny." Using a pre/post five-item scale [N = 15-25], we test whether players improve at spotting, questioning, and rewriting gendered fate-talk. We expect awareness and reframing confidence to rise [results pending data].
年轻人把算命的说辞当成对命运的中性描述,不知不觉就把里面的性别脚本内化了:同一张命盘,男人是「有魄力的领导者」,女人却是「太强势」「晚婚」乃至「克夫」。现有的偏见检测工具只盯着职场话术,AI 命理 App 只顾算得更准、从不教人反问——没有一个把「回嘴」这一步交到年轻读者手里。我们做了 Reframe Destiny(命运重构):一段短小的双语网页旅程——选八字或星盘,读一段传统的性别化解读,勾出你看到的偏见(婚姻中心论、克夫叙事、恐惧叙事)。随后并排对照一版 AI 重构的解读——它是受约束的教学脚本,不是真占卜——并在「命运法庭」写下自己的一句改写。我们用前测/后测的 5 题量表[N = 15-25],检验用户识别、质疑、重写命理性别话术的能力是否提升。我们预期偏见觉察与重构信心都会上升[数据待收集]。
AI-Bias Awareness Game · AI 偏见觉察游戏Cluster B
Most people treat ChatGPT or DeepSeek's advice as neutral, never noticing that the answer quietly assumes who is asking. Prior work audits bias inside the models but rarely lets an ordinary user feel it firsthand, and a classroom cannot run live audits. We built Pride and Prejudice in AI, a browser game where you ask the same question—say, competing for a promotion—through different simulated identities (gender, age, disability, region) and read the two AI replies side by side. Then, with identities hidden, you enter a blind room to judge which reply is more biased, and practice fairness-constraint prompts. Using a pre/post five-item awareness scale [N = 15-20], we track the change and blind-test accuracy; the compared replies are curated, not live. We expect awareness to rise and prompting to shift toward constraint [results pending data].
很多人把 ChatGPT、DeepSeek 给的建议当成中立答案,没意识到 AI 早已假设了「提问的是谁」。已有研究多在模型内部审计偏见,却很少让普通用户亲手感受它,课堂上也跑不了实时审计。我们做了《AI 中的傲慢与偏见》,一个网页小游戏:同一道题——比如「我想竞聘部门主管」——换不同身份(性别、年龄、残障、地域)去问,再把两条 AI 回复并排摆出来。接着进入盲测密室,在隐藏身份的情况下判断哪条更有偏见,并练习一套公平约束提问语。用一套前测/后测的五题觉察量表[N = 15-20],我们测觉察变化与盲测正确率;对比用的是受约束的策划回复,而非实时生成。我们预期体验后觉察上升、提问方式转向加约束[数据待补]。
Everyone Lines Up There · 大家都排那边(排队从众实验)Cluster C
When choosing between lines, people weigh a longer line as a signal that "this one is better" (social proof) against the simple logic that a shorter line is faster. We built Everyone Lines Up There, an online queue-choice simulation in which three lines are visually identical except for their length. Holding position and salience constant through randomization, we measure whether a visibly longer line raises the probability that high-school students choose it, and we code their open-ended reasons into reliability/conformity vs. speed cues [N = 15-20]. [Results pending: choice rate for the crowded line, distribution of reason categories, and whether stated reasons match actual clicks.] If the pattern holds, this shows that a lightweight simulation can surface how social signals override efficiency in ordinary decisions.
在两条队之间做选择时,人们把"更长的队"读成"这队更靠谱"的信号(社会证明),与"短队更快"这个朴素逻辑相互拉扯。我们做了《大家都排那边》——一个在线排队选择模拟:三条队伍除了长短,外观完全一样。通过随机化把位置和显著性都摁成一样,我们检验一条明显更长的队会不会抬高高中生选它的概率,并把他们的开放式理由编码为"靠谱/从众"与"更快"两类线索 [N = 15-20]。[数据待收:拥挤队伍被选中的比例、理由类别的分布、以及嘴上说的理由是否对得上实际的点击。] 若这一模式成立,它说明一个轻量的模拟就能让"社会信号如何压过效率"在日常决策里现形。

The confound control is written straight into the method sentence.

Random Identity · 随机身份(性别双标体验)Cluster C
Most people say they support gender equality, yet the same everyday action reads as "professional" for one gender and "not serious enough" for another. Prior work names this double standard but rarely lets an ordinary person feel it inside one small scene. We built Random Identity, a short browser experience where the player makes one identical choice—before a public speech, polish your appearance or rehearse your talk—first along one identity path, then the other. The comparison screen then says it plainly: same choice, different feedback. Using a within-subjects design with consented, anonymous play [N = 15-20], we drop the leading single-choice question and instead ask an open-ended prompt, coding answers into "noticed the double standard" versus "did not." We expect more players to name the everyday double standard after seeing both paths side by side [results pending data].
多数人嘴上支持性别平等,可同一个日常动作,换个性别就从「专业」变成「不够认真」。已有研究点出了这种双重标准,却很少让普通人在一个具体场景里亲身感到它。我们做了《随机身份》,一个几分钟的网页小体验:面对同一道选择——公开演讲前,是花时间打扮还是练稿——玩家先走一条身份线,再走另一条,最后落到对照页:同样的选择,不同的反馈。采用被试内设计、匿名且先取得知情同意 [N = 15-20];我们特意不用带引导性的单选题,而是改成开放题,把回答编码为「察觉到双标」与「未察觉」两类。我们预期,并排看过两条线之后,会有更多玩家说出这种日常双标 [results pending data]。
请按下面的「5 步模板」帮我起草一版我自己项目的占位 abstract(英文、学术、一段话):

1. Hook:现实痛点 +(大家常见的误解)
2. Gap:已有研究 / 工具还缺什么
3. Artifact:我做了什么,一句话 —【一句话描述你的作品】
4. Method:怎么测 —— pre/post + 大概 N + 因变量 —【你的前后测设计、大概多少人、量什么】
5. Expected:我预期会怎样

硬要求:
- 第 5 句只能是 "we expect…" 或者 [pending data],绝对不许出现任何结果数字(均值 / 百分比 / p 值都不行);
- 只用我给的信息,别编我没提供的细节,不够的地方用 [TODO: ...] 标出来让我补;
- 一段写完,别超过约 120 词。

06 · Write your Methods — today's real work

This is the real work of today — and it proves the point: you can write your whole Methods right now, because your tool is already built. Write it in the past tense, as something you already did — "players answered…", "the game logged…" — describing the version that actually shipped, not the plan from your midterm. Five blocks:

That honesty line matters more than it looks. The destiny-reframing and AI-bias projects both run on curated, scripted content, not a real API. Writing "the AI responses are a fixed, curated set built for this study" isn't a weakness you're owning up to — it's just being straight about your materials, and it makes the paper more believable, not less. Tell Cursor or codex your five blocks in your own words and have it write one clean Methods section:

请帮我写论文的 Methods(研究方法)一节,英文、学术、客观,用过去时 / 已实施的口吻写(我的工具已经上线了,不是计划):

- Artifact(作品):【描述你已经做好上线的东西:什么游戏 / 工具,玩家做什么,屏幕显示 / 记录什么】
- Design(设计):自变量、因变量、pre/post 前后测结构、比较的结果指标 —【填你的】
- Consent(知情同意):自愿、匿名、非临床,在第一屏呈现 —【填你的做法】
- Recruit + coding:招谁、大概多少人、怎么招;开放题怎么编码;跑什么统计 —【填你的】

硬要求:
1. 只根据我给的已上线版本写,别编我没提供的细节,不够的用 [TODO: ...] 标出来;
2. 如果我的 AI 内容是预设脚本 / curated mock(不是真实 API / 真实预测),照实写明这一材料边界 —— 这是加分项,别隐瞒;
3. 写完存进论文文件,别碰我的 Results / Discussion。
Today, no one touches the Results or Discussion prose. Lock Methods and you've done the two things that matter most: you unlock the method sentence of your abstract, and you turn Methods into a build / data-collection checklist for the week. Results is one placeholder line; Discussion waits for real data. Writing order today: Methods + placeholder abstract → then go collect → Results comes last.

07 · Cite it right, sound academic

Two things turn writing into a paper instead of a blog post: you cite your sources properly, and you write in an academic voice. Neither is hard once you've seen the pattern. No project here — just the rules.

First, citations — the APA basics. If you use someone's words, data, or idea and don't say where it came from, that's plagiarism (抄袭) — a red line, not a style slip. A citation has two halves: a short marker inside the sentence, and a full entry in the reference list.

APA — the two halves: In-text two forms, pick one: ... social proof shapes choices (Smith, 2021). Smith (2021) found that social proof shapes choices. References one entry, this exact shape: Author, A. A. (Year). Title of the article. Journal Name, Vol(Issue), pages. https://doi.org/xxx
AI invents fake references that look perfectly real. A plausible author, a real-sounding journal, a DOI that leads nowhere — you met this in earlier weeks. Every citation must be one you opened and confirmed exists. A fabricated reference is as fatal as a fabricated number.

Next, academic voice. Keep it objective, third-person, careful with what you claim. A few habits carry most of it — do the left, not the right:

08 · Fix these before you collect

Before you recruit a single person, run this checklist. Collecting on a broken design wastes real people's time and hands you data you can't use — so fix first, then collect. The blockers, in order of how many of you they catch:

① No pre-test = no story. Without a before measure you can never show the tool changed anyone's mind — the single most valuable sentence in your paper. Some of you even hid the pre-test; unhide it. This is fix number one.
② Consent first screen is missing. Most of your participants are minors — voluntary + anonymous + non-clinical on screen one is a hard gate before anyone plays.
③ "Supabase is connected" ≠ data is landing. Actually push one real row and confirm it's in the table. Don't trust the README.
④ Fix confounds first (Cluster C). A leaking design — unmatched line visuals, a leading question — gives you junk even at N = 100. Freeze the fix before collecting.
⑤ Narrow the main line (Cluster A). No locked unit / scope = no contribution sentence to write, so the abstract can't close.
⑥ Never fabricate. Real data ≈ 0 for everyone — write the placeholder, and never a number.

09 · Leave with — exit ticket

Everyone hands in the same three things before you leave — and the calendar is fixed: your data needs to start flowing ~7/7–7/10, or there's nothing to write next class.

10 · Before you go — class feedback

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