About the unverified marks on this pageFigures tagged unverified are PACE-local estimates that could not be confirmed against any published source, so they are labelled rather than quietly presented as fact. Every other factual claim on this page — statutes, standards codes, gene biology, primary-source citations — was independently checked; the full claim audit, including what came back wrong, lists every claim with its source.
Try it Tuesday · 90 minutes
Run this next week
90 minutes, one real standard. Minutes 0–15: name the grade level, the standard, and the specific learners. Minutes 15–50: run the differentiation prompt to generate four versions of one activity — below level, at level, above level, and an emerging multilingual learner — then evaluate each against the actual standard, rejecting what is subtly wrong for the grade. Minutes 50–90: with a partner, check the AI's most confident-looking choices — vocabulary, any translation, cognitive demand — and mark what a teacher must verify before using it. The deliverable is the evaluation, not the generated material; it grades professional judgment about AI output.
The full assignment
Make it 3D print · laser cut
A classroom manipulative designed for a specific partner teacher’s specific lesson. The constraint that makes it real: it has to survive thirty second-graders and cost almost nothing to reproduce.
Build it AI chat · no code
Differentiated versions of one lesson artifact, generated fast, then evaluated. Students see how quickly AI produces plausible-looking material and how much of it is subtly wrong for the grade level, which is the professional judgment being taught.
4 weeks, realistically
The 4-week version
- Wk 1Secure the partner classroom and the specific lesson. Students observe before designing, and cardboard-prototype the manipulative the same week so the fabricated part can be the second version, not the first.
- Wk 2Submit the refined manipulative fabrication file at the start of the week. Generate differentiated variants with AI, then evaluate against the actual standard and the actual learners. Most get rejected.
- Wk 3While the manipulative is fabricated, students finalize the lesson and rehearse teaching it.
- Wk 4Teach the lesson in the partner classroom. Debrief with the partner teacher, revise, hand it over.
Starter prompt & what the model actually did
The prompt, run for real
I am teaching this lesson to [grade level] students: [describe the lesson and standard]. Produce four versions of the main student activity, differentiated for: significantly below grade level, at grade level, above grade level, and an emerging multilingual learner. Keep the cognitive demand the same across all four; change the scaffolding, not the thinking. Then flag anything in your own versions that a teacher should check before using it.
Here's a single core activity — "Model the Journey of a Water Drop" — where students trace water through all four stages and explain what drives each change. The thinking (sequence + cause) stays constant; only the support changes.
**Core task (all four):** Show water moving through evaporation → condensation → precipitation → collection, and explain what makes each step happen (energy from the sun, cooling, etc.).
**Significantly below grade level**
Give 4 pre-made picture cards (sun heating a puddle, cloud forming, rain, lake). Students place them in order on a labeled arrow strip, then match each to a sentence starter: "The sun ___ the water, so it ___." Provide a word bank (heats, rises, cools, falls). They explain each step orally to a partner or you.
**At grade level**
Students draw their own 4-stage cycle diagram, label each stage, and write one sentence per stage explaining the cause ("The water evaporates *because* the sun heats it"). They finish with: "What would happen if the sun stopped shining?"
**Above grade level**
Same diagram and causal sentences, plus: predict what changes if one variable shifts (hotter day, colder air) and justify it. Add a challenge — trace the *same* water molecule through two full cycles, and explain why the cycle has no true beginning or end.
**Emerging multilingual learner (Ilocano L1)**
Same 4-stage diagram as the at-grade version, with these supports: bilingual labels (English + Ilocano — e.g., *danum* = water, *ulep* = cloud, *tudo* = rain, *init* = sun), the picture cards from the below-level version, and sentence frames with a word bank. Allow first draft in Ilocano or drawing; the science reasoning is assessed, not English fluency. Pair with a supportive partner if one shares the language.
**Flag these before using — teacher checks:**
1. **Verify the Ilocano vocabulary.** I generated *danum, ulep, tudo, init* from general knowledge — confirm spelling and word choice with a fluent speaker, family member, or your EL coordinator before printing. Do not trust my translations unchecked.
2. **Confirm the NGSS alignment.** Hawaiʻi adopted NGSS, but the water cycle as a named 4-stage cycle isn't cleanly a single grade-3 performance expectation — much of it lives in grade 2 (2-ESS2-3, water in solid/liquid form) and grade 5 (5-ESS2-1, Earth's systems). Check your grade-3 pacing guide/standard code so the lesson maps to what your school expects.
3. **Watch the vocabulary load.** "Evaporation/condensation/precipitation/collection" are demanding for all four groups, not just the multilingual learner — decide whether you're teaching the terms or the concepts first.
4. **The "same molecule, two cycles" extension** assumes conservation-of-matter reasoning some 3rd graders won't have yet; preview it before assigning.
5. **Cultural relevance:** consider grounding examples in local water (rain on the mountains, streams, ocean) rather than generic ones — strong for all learners, especially in Hawaiʻi.
Want me to build the printable picture cards or a rubric that scores the reasoning consistently across all four versions?
- Flagged its own Ilocano translations: 'Do not trust my translations unchecked'The model refused to vouch for danum/ulep/tudo/init and told the teacher to verify with a fluent speaker — the exact human-check the assignment teaches.
- Corrected the standard alignment: water cycle 'isn't cleanly a single grade-3 performance expectation'It caught that the prompt's NGSS framing was off (the content spans grades 2 and 5) — professional-grade scrutiny of the user's premise.
- Held cognitive demand constant across versionsDifferentiated the scaffolding, not the thinking, as instructed — the design principle the rubric rewards.
- We checked the Ilocano: all four words are correctdanum (water), ulep (cloud), tudo (rain) and init (sun) are all attested Ilocano — tudo is the root behind agtudo, "to rain," and init is the standard entry for sun. Ilocano also has uran for rain and aldaw for sun, so a fluent speaker may still prefer a different register for a specific classroom. Worth sitting with: the model was right, and it was still right to refuse to vouch for itself. The rubric grades the verification, not the outcome — a teacher who printed these unchecked got lucky, and would have no way of knowing which time was the unlucky one.
Assessment
Rubric
| Criterion | Weight | What it assesses |
|---|---|---|
| Standard alignment | 25% | Each version is checked against the actual grade-level standard, and any misalignment the AI introduced is caught. |
| Constant cognitive demand | 25% | Differentiation changes scaffolding, not the thinking; the student verifies the AI did not lower the demand. |
| Learner-specific scrutiny | 30% | Flags what must be verified for real learners — translations, cultural references, reading load — rather than trusting the output. |
| Rejection judgment | 20% | Rejects the versions that are wrong for the grade and explains why, rather than accepting plausible material. |
What this replaces
Swapping it into a real course
Replaces
The lesson-plan submission graded on format and completeness.
What is lost
The from-scratch lesson-writing practice.
What is gained
Students practice the judgment that now matters most — spotting where fast, plausible AI material is wrong for these learners and this standard — instead of being graded on a template the AI can fill in seconds.
Where AI is bad at this
The failure your students should catch
A model produces differentiated lesson material that looks classroom-ready and is subtly wrong for the grade — misjudging cognitive load, mislabeling a standard, and generating confident 'translations' or cultural references it cannot vouch for. It changes the thinking when asked only to change the scaffolding, and it never flags that a human must check the pieces most likely to mislead a specific learner.
Budget & logistics
What it costs to run
- Instructor prep1.5 hours
- Class time90 minutes
- Per-student cost$0 for the Tuesday version; roughly $4–10 per student in filament or cut stock unverified for a durable manipulative in the four-week build.
- Fabrication file dueFirst day of Week 2 — cardboard-prototype the manipulative during week-1 observation, then submit the refined fabrication file at the start of week 2 (still the second version, iterated earlier). PACE quotes 7–10 business days (up to 14 calendar days), so it clears before week-4 classroom teaching.
- Calendar dependencyMaterial generation and evaluation need no lead time. The manipulative is for the specific lesson secured in week 1; students cardboard-prototype it that week and submit the refined file at the start of week 2, so the up-to-14-calendar-day turnaround clears before week-4 teaching.
Three sizes
Scale it to the time you have
One session
The 90-minute generate-and-evaluate session producing an evaluation of the AI material; no fabrication.
4 weeks
The seeded four-week plan: secure a partner classroom, generate and vet differentiated variants, prototype then fabricate the manipulative, then teach and revise with the partner teacher.
One semester
A partner-classroom project — observation, AI-differentiated materials vetted against standards, a prototyped-then-fabricated manipulative, taught in the real classroom and revised with the partner teacher.
Reserved · live build
This space is intentionally empty. During the Day 8 session it will be filled in live — fill the liveBuild field in content/teacher-education.json and rebuild.
Tell us what happened
Run it, then say how it went
If you try this — the 90-minute version, the prompt, any part of it — send back what you tried and what happened, especially anywhere the model was confidently wrong. That is the material the next session is built from.
Email feedback on Teacher Education →
Opens a draft in your mail client, already addressed and titled. No form, no account, nothing to sign up for.