Learning Hub / Teacher kit
Teacher kit
You check that every learner understood the task, help one sealed forecast happen in class, and make sure each learner can explain one error or spot one unsupported claim. The program does the scoring. Nobody ranks children.
你负责确认每个孩子读懂题目,课堂内完成一次封存,并让每个孩子能解释一个误差或指出一个无根据的推断。算分交给程序,不给孩子排名。
One-page handbook
You are responsible for
- Checking the task is understood: station, day, cutoff.
- One sealed forecast in class, with a receipt.
- At review time, each learner explains one error or names one unsupported claim.
- 确认孩子读懂题:站、日、截止。
- 课堂内完成一次封存并拿到回执。
- 复盘时每个孩子说出一个误差的解释或一个无根据的推断。
You are not responsible for
- Scoring (the program does it, from the official report).
- Ranking, or judging who is "talented".
- 算分(程序依据官方报告计算)。
- 排名,或判断「谁更有天赋」。
Three actions every round
- Opening day: project the current round; confirm everyone has a receipt (check the first 8 characters of the sha256).
- Reveal day: project the observation; each learner computes their own error; compare with the program's result.
- Record: completion rate, teacher minutes, one question.
- 开题日:投影当前轮次;确认每人拿到回执(看 sha256 前 8 位)。
- 揭晓日:投影实测;每人算自己的误差;与程序结果对照。
- 记录:完成率、教师分钟数、一个问题。
Exceptions. Data PENDING: teach "we wait honestly". A learner wants to change a forecast after the cutoff: a revision is a new version, it never replaces the scored one. Red lines: never share a child's information, never post children's photos publicly, never submit for a child.
异常处理。数据 PENDING:讲「如实等待」。孩子想改已截止的预测:修订是新版本,不替换计分版本。红线:不向任何人透露孩子信息;不在公开区发孩子照片;不替孩子提交。
Child task card (one per round, printable)
- My station: Boston Logan airport (the same for everyone).
- My climate day: the start and end shown on the round (fixed EST 00:00–24:00; the page also shows my local time and UTC).
- Latest time to submit: 18:00 EST the day before. It must be before; exactly at the cutoff counts as late.
- My forecast: high temperature ____ °C (one decimal).
- What I used: public weather information / my own calculation / AI help / other (tick what is true; using AI neither adds nor removes points).
- My one-line reason: ____ (learning record only, does not affect the weather score).
- Press seal and keep the receipt. Changing before the cutoff makes a new version; the old one is kept.
- After the reveal: observed ____, error ____; I want to check ____; next time I will change only ____.
- 我预测的站:Boston Logan 机场站(所有人同一个)。
- 我预测的观测日:照题目显示的开始与结束时间(固定 EST 00:00–24:00;页面另显示我的本地时间与 UTC)。
- 最晚什么时候交:前一日 18:00 EST,必须在这之前,恰好等于也算晚。
- 我的预测:最高温 ____ °C(一位小数)。
- 我参考了什么:公开天气信息 / 自己计算 / AI 辅助 / 其他(按事实选;用了 AI 就勾,不扣分不加分)。
- 我这样猜的一句话理由:____(只进学习记录,不影响天气分)。
- 按「封存」后拿到回执;截止前改,就是新版本,旧版留着。
- 揭晓后:看实测 ____、误差 ____;我想核查 ____;下一次只改变 ____。
Teacher judging card (each round)
- First check the receipt time is before the cutoff; then check source, station, unit, time zone and quality flags. If incomplete, wait; never fill in zero.
- After program scoring, spot-check by hand against the original: 57 °F = 125/9 °C; error = |9t − 50F + 1600| / 90. Try it in the exact error calculator.
- Models, teachers and AI have no way to override a score. In a dispute the original result stays; a review record and a corrected version are added.
- Make one judgment per session: can the learner explain one error, or spot one unsupported inference? If similar mistakes repeat, check the question, data and program before deciding it is a learning gap.
- Never reveal a child's information, never post photos publicly, never submit for a child.
- 先核回执时间在截止前;再核源、站、单位、时区和质量标记;不完整先等,不能填零。
- 程序评分后抽查手算与原件:57 °F = 125/9 °C;误差 = |9t − 50F + 1600| / 90。可用精确误差计算器核对。
- 模型、教师、AI 都没有覆盖成绩的入口;争议保留原结果,新增审核记录与更正版。
- 每次学习只作一个判断:孩子能否解释一个误差,或识别一个无根据的推断。连续相似错误先查题面、数据与程序,再判断学习断点。
- 不向任何人透露孩子信息;不在公开区发照片;不替孩子提交。
First activity: 20 minutes in two short touches
| 3 min | Look at Logan station and the climate day; learners say what is being forecast.看指定 Logan 站与观测日,学生说清「预测的是什么」。 |
| 5 min | Look at public data or a historical sample; estimate a high first.看教师提供的公开数据或历史样本,先估一个最高温。 |
| 5 min | Enter the forecast and an optional reason; get the receipt; teacher checks the cutoff.填预测、可选一句理由,得到回执;教师核对截止。 |
| 5 min | After the reveal (separate touch): observed value, absolute error, one method to change.揭晓后另用 5 分钟:看实测、算绝对误差、说一个想改变的方法。 |
| 2 min | Guardian sees a private summary and confirms whether to continue.家长看到私密摘要,确认下一次是否继续。 |
Historical practice and real before-the-fact forecasts must be labelled separately.
Five numbers (every round, write zero as zero)
Real forecasts submitted early · valid observations · learners who got feedback · teacher minutes per learner · learners willing to join next time. Also note data delays and reasons for leaving. Never use page views or synthetic samples in place of participation. The card below stays in this browser only.
真实提前提交数 · 有效观测数 · 拿到反馈数 · 每人教师分钟数 · 下一次愿意参加数。另记数据延迟与退出原因。不用页面浏览或合成样本代替参与。下方卡片只保存在本浏览器。
Reading-card template
Each card must be completely filled before use. A card for a source nobody has read in the original may only be marked "candidate" and does not enter the course pack.
每张卡必须填满才可用。未读原文的卡只允许写「候选」,不进课程包。
Card no.: RC-0x
Source / chapter: __ (real source path + version or commit anchor + page/paragraph)
Age band: __
Key points from the original (max 120 words, cite page): __
Teacher rewrite (3 sentences for children): __
Three comprehension questions + answers: 1.__ 2.__ 3.__
Common misconception check: __
Matching task (max 10 minutes): __
Approved by / date: __
卡号:RC-0x
书线/章节:__(真实源路径 + 版本/提交锚 + 页/段)
适用年龄段:__
原文要点(≤120 字,引用标页):__
老师改写(给孩子的 3 句):__
三道理解题 + 答案:1.__ 2.__ 3.__
常见误区检查:__
配套任务(≤10 分钟):__
核准人/日期:__
Teacher workshop: run one round in 90 minutes
Everything in the workshop is SYNTHETIC (workshop pack W1) and never enters a real round. Groups of 3–4 teachers rotate roles: learner, receiver, revealer, checker. Never rehearse with real children's data.
工作坊全部为合成资料(工作坊包 W1),不进入正式轮次。每 3–4 位教师一组,轮流担任学习者、签收员、揭晓员、复核员。不要用真实儿童资料排练。
Card W1 (SYNTHETIC): Boston Logan USW00014739; TMAX in °C; three days 2026-12-02 to 12-04 sealed at once; fixed EST climate days; cutoff 2026-12-01 18:00 EST = 23:00 UTC; baseline = last complete public day 8 °C for all three days. Versions: V1 16:20 [8, 10, 10]; V2 17:55 [9, 12, 11]; V3 12-05 03:00 [12, 14, 13] (after the results). Reveal [12, 14, 13].
题卡 W1(合成):Boston Logan USW00014739;TMAX,°C;2026-12-02 至 12-04 三日一次封存;固定 EST 气候日;截止 2026-12-01 18:00 EST = 23:00 UTC;基准为截止前已公开的上一完整日 8 °C,三天均 8 °C。版本:V1 16:20 [8, 10, 10];V2 17:55 [9, 12, 11];V3 12-05 03:00 [12, 14, 13](结果已出)。揭晓 [12, 14, 13]。
| 0–10 | Start from the rules: which sentence could still be read two ways? Each group fixes one ambiguity.从规则开始:哪一句仍可能有两种解释?每组指出一项歧义并更正。 |
| 10–22 | Experience sealing: V1, V2 and sign-off cards.体验封存:V1、V2 及签收卡。 |
| 22–32 | Late and clock-rollback cards: learning revisions are kept, the scored version is not replaced.晚交/回拨卡:学习修订保留,正式版不覆盖。 |
| 32–47 | Synthetic reveal, hand check: absolute errors first, then the average.合成揭晓、手算复核:先逐日绝对误差,再平均。 |
| 47–62 | Incident table-top: each group draws two situation cards.故障桌面演练:每组抽两张情况卡。 |
| 62–75 | Feedback and a 60-second parent explanation: talk about evidence and next steps, never "gifted" or "not suited".反馈与 60 秒家长说明:说证据和下一步,不说天才或不适合。 |
| 75–85 | Five-minute teach-back using one question from Unit 2.五分钟带教:用第 2 单元的一题教同伴。 |
| 85–90 | Decide the scope: promise only what you have tested; if the core rehearsal failed, start with paper.决定开课范围:只承诺已实测可用的功能;核心演练未过则纸笔课先行。 |
Workshop answer key
All three versions are kept; V2 is scored. Learner errors [3, 2, 2], MAE 7/3 ≈ 2.33 °C; baseline errors [4, 6, 5], MAE 5 °C; on this sample the improvement is 8/3 ≈ 2.67 °C. Three samples cannot support "significantly better long-term ability".
三份都保留,V2 计分。学生误差 [3, 2, 2],MAE 7/3 ≈ 2.33 °C;基准误差 [4, 6, 5],MAE 5 °C;本样本改善 8/3 ≈ 2.67 °C。三个样本不能支持「长期能力显著提升」。
Situation cards
| Situation | Reasonable handling | Not acceptable |
|---|---|---|
| Official observation delayed | Mark PENDING | Quietly fill in |
| Target day has no observation | No weather error for that event | Treat as 0 °C |
| Observation corrected | Keep both and announce | Overwrite silently |
| Learner's device offline | Paper continues, labelled | Fake a receipt |
| Revised after seeing the result | Keep as late learning version | Overwrite |
| Learner needs input support | Verbatim, recorded | Hints as low ability |
| Parent declines publication | Private learning continues | Penalize |
| A real address appears in a reason | Restrict at once | Use in publicity |
Feedback you can say directly
"You kept your original forecast and can explain why V2 counts. Your errors were 3, 2 and 2 °C. Saying you ran low overall this time is supported; whether you run low again needs a new task. Next time write one reason before the cutoff and note when your data was published."
「你保留了原预测,也能解释为什么 V2 计分。三天误差是 3、2、2 °C。你说这一次整体偏低有证据;下一次是否仍偏低还需要新任务检验。下次请在截止前写一句理由,并标资料发布时间。」
Ready to teach when: two teachers independently get the same V2 and MAE, handle a missing observation, a late version and a "do not publish" correctly, and the task card has no ambiguity. A passed rehearsal proves the rehearsal, not that a production system is safe.
准备就绪判定:两位教师独立得到同一个 V2 和 MAE;能正确处理缺观测、迟交与不公开三种情况;题卡无口径歧义。一次排练通过只证明排练流程,不证明生产系统已安全上线。
Sources: spec 43, spec 45, m681 TEACHER_WORKSHOP.