The 42nd Period | STEM Education Hub

🎯The Challenge

read this first

"AI makes me more productive" is a claim. In 2.1.2 you tested it on a task someone else designed. Now you design the test. You pick three real tasks, decide what a good result looks like before you start, do each task by hand and with AI, and count what each way cost, including prompting, waiting, checking and fixing. A partner scores the results without knowing which one the AI made. The product is a Productivity Study Memo with your numbers, a verdict for each task, a chart, and a recommendation you can defend.

Rules of the Study

Rule first. Before any task, check the rule that applies to you (2.1.3). If the rule says not allowed, pick a different task. If it doesn't cover your use, ask, and write down the answer.
Real, small, and safe. Choose tasks you would really do that take about 20 minutes or less by hand. Don't use graded assignments unless your teacher says you may.
No private details in prompts. No names, addresses, or anyone else's information. Use your own notes, invented situations, or public material.
Write the criteria before you run. Otherwise you'll grade the results after you've seen them.
Time honestly. A rough number is fine. Time the prompt, the wait, the checking and the fixing separately.
Report what you found, even if the AI didn't help. "It didn't help" is a good result if the evidence is clear.

0 of 26 checkpoints done · your work saves in this browser as you type

🧰Study Toolkit

everything new in this project

Most of what you need you already learned: measuring both ways and the order effect (2.1.2), checking against the rule (2.1.3), and fit versus truth (2.1.1). These are the few new tools. Open each one when you reach that step.

1 · A fair test

Same task, both ways. The by-hand version uses you, a calculator or a spreadsheet, and no AI. Cap it at about 20 minutes.
Vary the order. Do one task by hand first, and another AI first. The second try is always easier, as you saw in 2.1.2. Mixing the order doesn't remove that, but it lets you see whether it matters.
Similar conditions. Same place, no interruptions, and about the same time of day if you can. Write down anything unusual.
Pilot first. Do a 3-minute task both ways just to practice timing the four parts. Fix your routine before it counts.
Check it yourself. A task with a quick check (compare to a key) and one with a slow check (redo the work) will behave differently. You need both.

2 · Decide what good looks like first

Write four yes/no criteria for each task before you start. Each should be something a stranger could check: "contains my 5 key points," "under 100 words," "matches the key."
Avoid "sounds good," "is helpful" or "is well written." Two people won't agree on those.
Score blind. Flip a coin to label the by-hand result X or Y (and the AI result the other letter). Put both in one document. Your partner scores each one against your criteria without knowing which is which, and you reveal it afterward.
Your score for each version is how many of the four criteria it meets, from 0 to 4. A difference of 1 point counts as better or worse. With four criteria, that is a coarse measure, and the memo should say so.
Score the AI's first draft too (you can do this one yourself). The gap between the first draft and the final version shows what the checking and fixing bought.

3 · Counting and verdicts

AI total = prompt + wait + check + fix. Minutes saved = by hand − AI total. Within 10% of your by-hand time counts as about even.
Quality: the AI's score after checking and fixing, compared with your by-hand score. One point or more is better or worse.

Time ↓ / Quality →	Better	Same	Worse
AI saved time	Helped	Helped	Mixed
about even on time	Helped	No clear difference	Didn't help
AI cost more time	Mixed	Didn't help	Didn't help

"Mixed" means one thing improved and the other got worse. The page works out the verdict. Your job is to explain it.

4 · Chart it honestly

The title says what the chart shows (not just "Results").
The vertical axis says "minutes." The horizontal axis names the tasks.
The bars start at zero.
A legend (or labels) says which bar is by hand and which is the AI total.
Every bar matches a number in my table. I checked each one.
Both bars for a task use the same scale and the same units.
One sentence under the chart says what it shows.

Build your own chart first. Then ask an AI to make the same chart from your table (paste numbers only), and run the same checklist on its chart. Charts from a tool can look polished and still have a wrong bar, a missing label, or a baseline that doesn't start at zero.

5 · Say what three tasks can't tell you

Small sample. Three tasks is three examples, not a rule about AI.
One person. The result depends on how good you already are at the task, and how good you are at prompting.
One tool, one day. A different tool, or the same tool tomorrow, may do differently. Re-running an AI version shows how much it varies.
Order and learning. Whatever you did second had help from the first.
Scoring. Four criteria is a coarse scale, and your partner's reading of a criterion is one reading.
A memo that names these honestly is stronger than one that claims a big result.

6 · Two invented examples, worked out

Task	By hand	AI: P + W + C + F	Saved	Quality (hand → draft → final)	Verdict
Summarize a 600-word article into 5 bullets	14 min	2+1+7+2 = 12	2	3 → 2 → 4	Helped (AI saved time, better quality)
Make a 5-day study schedule that follows 4 rules	10 min	3+1+9+3 = 16	-6	4 → 3 → 4	Didn't help (AI cost more time, same quality)

In the first row, checking and fixing took 9 of the 12 AI minutes. People who say "it took 3 minutes" count only the prompt and the wait. In the second, the AI draft was fine only after 12 minutes of checking and fixing, and by hand was faster with the same quality.

🤝0 · Partner Up

Day 1 · 10 min

You each run your own study. You are also each other's blind scorer, so agree on how you'll work.

My name (or initials)

My partner (the blind scorer)

Chatbot tool I'm using, and today's date

How and when will we swap results for scoring?

What will we NOT tell each other until scoring is done?

What do we do if one of us falls behind?

You and your partner wrote down three working norms.You read the Rules of the Study below, and you agree to follow them.

🧭1–2 · Choose and Clear Three Tasks and Design the Test

Day 1 · 35 min

Pick three real tasks of three different kinds, and check each against the rule that applies to you. Then decide what good looks like and what order you'll work in before you start anything.

Task 1

What is the task?

What kind of task is it?

The rule that applies to this use (from 2.1.3)The exact words of the rule (or the answer you were given), and where you found them

How hard would it be to check the AI's result?

How exactly will you check it?

Which version will you do first?

Four yes/no criteria, written before you run either version

Flip once. It decides which letter goes on the by-hand result.

Task 2

What is the task?

What kind of task is it?

The rule that applies to this use (from 2.1.3)The exact words of the rule (or the answer you were given), and where you found them

How hard would it be to check the AI's result?

How exactly will you check it?

Which version will you do first?

Four yes/no criteria, written before you run either version

Flip once. It decides which letter goes on the by-hand result.

Task 3

What is the task?

What kind of task is it?

The rule that applies to this use (from 2.1.3)The exact words of the rule (or the answer you were given), and where you found them

How hard would it be to check the AI's result?

How exactly will you check it?

Which version will you do first?

Four yes/no criteria, written before you run either version

Flip once. It decides which letter goes on the by-hand result.

I wrote every criterion above before doing either version of any task.

Each task is something I would really do, and it is small enough to do by hand in about 20 minutes or less.I checked the rule for each task, quoted its exact words with the source, and asked when the rule didn't cover it.One task is quick to check and one is slow to check.My teacher approved my three tasks.

I wrote 4 yes/no criteria for each task before doing any version of it.I chose an order for each task so both orders appear.I flipped the coin for each task, and I haven't told my partner the result.I did a 3-minute pilot (a tiny task both ways) to practice timing the four parts of the AI way.

⏱️3 · Run Both Ways

Days 2–3

Do each task both ways, timing the parts, under similar conditions. Whole minutes are fine.

Task 1: (name it in the plan)

Planned order: not chosen yet

By hand (min)

AI: prompt (min)

AI: wait (min)

AI: check (min)

AI: fix (min)

Both versions were done in similar conditions (same place, no interruptions)

Notes and anything unusual (an interruption, a bad prompt, the AI changing its answer)

Task 2: (name it in the plan)

Planned order: not chosen yet

By hand (min)

AI: prompt (min)

AI: wait (min)

AI: check (min)

AI: fix (min)

Both versions were done in similar conditions (same place, no interruptions)

Notes and anything unusual (an interruption, a bad prompt, the AI changing its answer)

Task 3: (name it in the plan)

Planned order: not chosen yet

By hand (min)

AI: prompt (min)

AI: wait (min)

AI: check (min)

AI: fix (min)

Both versions were done in similar conditions (same place, no interruptions)

Notes and anything unusual (an interruption, a bad prompt, the AI changing its answer)

I timed the prompt, the wait, the checking and the fixing separately for every AI version.I did both versions of each task in similar conditions (same place, no interruptions), or I noted when I didn't.I saved the AI's first draft before fixing it.I re-ran at least one AI version a second time and kept its result.

🕶️4 · Score Blind

Day 3 · 15 min

Your partner scores each pair against your criteria without knowing which was the AI's. Scores are out of 4: one point for each criterion met.

Task 1: (name it in the plan)

By hand (partner scored)

AI first draft (I scored)

AI after fixes (partner scored)

AI second run (optional)

My partner scored the X and Y copies for this task without knowing which was the AI's.

Task 2: (name it in the plan)

By hand (partner scored)

AI first draft (I scored)

AI after fixes (partner scored)

AI second run (optional)

My partner scored the X and Y copies for this task without knowing which was the AI's.

Task 3: (name it in the plan)

By hand (partner scored)

AI first draft (I scored)

AI after fixes (partner scored)

AI second run (optional)

My partner scored the X and Y copies for this task without knowing which was the AI's.

For each task I put the two results in one document, labeled X and Y by the coin flip.My partner scored X and Y against my criteria. I told them nothing about which was which.I scored the AI's first draft against the same criteria myself.I entered the scores below.

📊5 · Findings and Chart

Day 3 · 20 min

Check the page's numbers against your log, then make a chart from your table. Everything in the table is computed from your log.

Task	By hand	AI total	Saved	"Reported" (P+W)	Check + fix	Quality (hand → AI)	Verdict
Task 1	Fill in every minute and score for this task.
Task 2	Fill in every minute and score for this task.
Task 3	Fill in every minute and score for this task.

Data table for your chart

Your chart table appears here once a task is complete.

Chart checklist: run it on your chart, then on the AI's chart

The title says what the chart shows (not just "Results").The vertical axis says "minutes." The horizontal axis names the tasks.The bars start at zero.A legend (or labels) says which bar is by hand and which is the AI total.Every bar matches a number in my table. I checked each one.Both bars for a task use the same scale and the same units.One sentence under the chart says what it shows.

When you asked an AI for the same chart from your table, what did it get right, and what did it get wrong or leave out?

I checked the page's totals against my own log.I built a bar chart of minutes (by hand vs. AI total), in a spreadsheet or on paper, and checked it against the list.I asked an AI for the same chart from my table (numbers only), and compared it with mine.

🧾6 · Memo

Day 4 · 25 min

Write the Productivity Study Memo: the numbers, a verdict for each task, and a recommendation that fits what you found. The method and results fill themselves in from your work.

Your recommendation: when will you use AI for work like this, and when won't you? Give reasons from your own numbers and the three questions (Can I check it faster than I could do it? What if it's wrong and I miss it? Am I supposed to be learning this?)

What can't this study tell you? Think about sample size, order, scoring, and the tool (0 of at least 15 words)

AI-use statement (from 2.1.3)

Tool

Used for

Not used for (I did this myself)

What I checked, and how

Policy that applies (source and the words)

Checks (computed from your work)

Three tasks, each with a name and a kind. 0 of 3 tasks have a name and a kind.

Three different kinds of task. Pick a different kind for each task, so your results aren't one result three times.

Every task is cleared by the rule that applies. 0 of 3 tasks cleared. Each needs an answer and the exact words with the source.

4 yes/no criteria per task, written before the run. 0 of 3 tasks have 4 criteria.

The order varies: at least one hand-first task and one AI-first task. Choose an order for each task, and make sure both orders appear (the order effect from 2.1.2).

One task is quick to check and one is slow. Mark how each task would be checked, and include at least one quick check and one slow check.

All three runs are complete. 0 of 3 tasks have every minute and score filled in.

A partner scored each pair without knowing which was the AI's. Flip the coin for each task, then tick each task's box in Score once your partner has scored the X and Y copies.

Chart built and checked (5 of 7 items), plus the AI-chart comparison. 0 of 7 items ticked, and the AI-chart comparison is empty.

Recommendation and limits are written. Still needed: your recommendation, limits (0 of 15 words).

AI-use statement is complete. Still empty: tool, used for, not used for, what you checked, policy that applies.

Your memo · not finished yet

PRODUCTIVITY STUDY MEMO
By: (name)   Partner who scored blind: (partner)
Tool: (tool and date)

QUESTION: For three everyday tasks, did using AI save time and keep quality, once checking and fixing are counted?

METHOD
1. (task) [kind?]. Order: ?. Check: ?.
   Rule that applies: ?. "(missing)"
   Quality criteria, written before the run: (missing)
2. (task) [kind?]. Order: ?. Check: ?.
   Rule that applies: ?. "(missing)"
   Quality criteria, written before the run: (missing)
3. (task) [kind?]. Order: ?. Check: ?.
   Rule that applies: ?. "(missing)"
   Quality criteria, written before the run: (missing)
Scoring: my partner scored the by-hand and AI results blind (labeled X and Y by a coin flip), out of 4.

RESULTS (minutes; AI total = prompt + wait + check + fix)
1. Task 1: (run incomplete)
2. Task 2: (run incomplete)
3. Task 3: (run incomplete)

CHART: 0 of 7 chart checks passed. AI-chart comparison: (missing)

RECOMMENDATION: (missing)

LIMITS: 3 tasks, one person, one tool and a few days is a small study, so it can't show what AI does in general. (missing)

AI-USE STATEMENT
  Tool: (missing)
  Used for: (missing)
  Not used for (I did this myself): (missing)
  What I checked, and how: (missing)
  Policy that applies: (missing)

Add your chart (a screenshot or your spreadsheet) when you hand it in.

Every check on this page is green or explained.My recommendation says when I'll use AI and when I won't, with a reason from my own numbers.I copied the finished memo and handed it in the way my teacher asked.

🗣️7 · Share

Day 4 · 20 min

Present your data, your chart, how you used AI, and what you learned.

Your 3 minutes:

The data and the chart. One task in detail, with the four AI times.
How you used AI. What you asked it, what you checked, and what you fixed.
Your key insight. One thing the numbers showed that you didn't expect, or that matched your prediction.
Your recommendation, and one limit.

When you listen to someone else, ask:

Can I find that number in their log? Did they count all four parts of the AI time?
Were their criteria written before they ran the task? Who scored, and did the scorer know which was the AI's?
What does their data not show? Would a different task or tool change the result?
Did the AI's chart match the real numbers?

I presented my data and chart, how I used AI, and my key insight (3 minutes).I gave another student one strong point and one question about their evidence.

📐What Strong Work Looks Like

Part	Meets the project	Strong
Tasks	Three real, different tasks, each cleared against the rule with its exact words.	Tasks are ones you would really do, and the quick-check and slow-check tasks are chosen to teach something.
Design	Four yes/no criteria per task written before the run; order varied; coin flipped.	Criteria are checkable by a stranger, and you can say how each one could have been unfair.
Measurement	All four AI times and all scores recorded for every task.	You noticed the checking and fixing share, and you re-ran an AI version to see how much it varies.
Scoring	A partner scored blind.	You can describe a disagreement over a criterion and how you settled it.
Chart	Bars start at zero, labeled, and match the table; the AI's chart is compared.	You found a specific thing the AI's chart got wrong or left out.
Memo	Verdicts for each task, a recommendation tied to your numbers.	The recommendation uses the three questions and says what would change your mind.
Honesty	Limits are written and include at least two real ones.	You say what three tasks can't show, with no claims about "AI" in general.

🛠 Project 2.1.4 · Unit 2 of 3 · 11 Weeks