This piece has an evidence version. The same subject was made again, with running code and screens that were actually rendered — The Standard Stays Yours — What a Grading Tool Does Not Do. This one is left as it was written.
A single red pen lay on the desk of Korean teacher Han Mi-jeong (41). Even though it was the busy week when performance assessments pile up, there was no stack of answer sheets beside her. Instead a tablet stood propped at an angle, and on the screen sat a list titled "today's answers to grade" — student numbers, items, and small check boxes alongside. She swiped through the screen once and said, "Just one class left today."
Four classes, 132 students. Essay-style performance assessments come around weekly. Usually around this time she'd stack the answer sheets on her dining table and grade past midnight. Yet her evening was empty. Asked her secret, she turned the tablet around. "This screen. I made it myself." She said she'd never learned to code.
How many answer sheets do you grade in a semester?
I've never counted… but with 132 students and five or six assessments a semester, that's a hundred and thirty sheets at a time, several times over. And they're essays. With multiple choice you can stack the sheets and grade them together, but essays have to be read one by one.
The problem wasn't the volume — it was the repetition. Honestly, reading one sheet and deciding a score is fast. A minute will do. The criteria are sharp in my head. For this item: two key concepts and one appropriate example earn full marks, one concept alone earns half, a wrong example is a deduction, concepts that are written but contradict each other get yet another treatment. It's a yardstick I refined over twenty years of grading, across thousands of answers. It isn't written in any textbook. So I can't just hand the whole thing off to someone else. This standard exists only in me.
But recording that judgment eats the whole evening. The evening after a test, I'd pile the answers on the dining table and open the first sheet. Read one, apply the criteria in my head, mark the score, jot a note in red pen, copy it into the grading sheet's box, next sheet. And the next. The judgment takes a minute, but writing that judgment down by hand, adding it up, transcribing it — that takes several times longer. Moving the same criteria by hand a hundred and thirty times. By the time midnight nears, I have no energy left to read the answers deeply. My hands are busy but my head goes foggy. The attention I should be spending on grading leaks out into the manual work.
So you built your own tool. Did you learn to develop?
No. That's the whole point. I can't write code. All I did was decide what was needed. Which item needs which grading criteria, what earns full marks and what gets deducted, what my hands keep repeating while I grade — I know that because I've taught Korean for twenty years. Translating it into something a computer understands was always the obstacle, and makemind took over that part.
I didn't draw the screen — I spoke it. When I pick an item, let that item's grading criteria appear as a checklist. When I tap the items that are met, let the score add up automatically. A line for a note beside it. At the top, count how many I've done. When you write down what to show and where like this, that becomes the screen. I didn't design anything — I just said what I needed.
"I didn't draw the screen, I spoke it. Because I know best what matters. I may not be able to draw, but I can speak."
Can you show me what actually happens behind that screen?
Of course. The truth is the screen comes last. There are a few stages before it. My grading criteria get organized into items, the answers come in, similar answers are grouped so I see them by the same standard, and a grading history builds up per student. I didn't wire that flow — I only decided what goes in and what comes out of each stage.
The first stage — how do the grading criteria go in?
I just wrote down what had been in my head as items. "Key concept A," "Key concept B," "Appropriate example," "Deduction: faulty example," like that. With a score attached to each item. Before, these criteria lived only in my head. So they were hard to explain to other teachers, and even I wasn't sure whether I was seeing the hundredth sheet the same way as the first.
Once I wrote it out as items, that became the skeleton of the grading screen as is. I didn't build a screen separately. I wrote down the criteria, and that became the screen. If I'd told an outside developer "build me a grading app," they'd have had to interrogate me at length about my criteria first, and something subtle would have leaked out in the process. "What's a key concept?" "By what standard is an example appropriate?" — could you really put all of that into words? Because the person who knows the criteria moves them over directly, nothing leaks. That's eighty percent of the tool. The missing twenty percent was just the screen axis to hold those items, and that's something you lay on top.
How do the answers go in, and what does "grouping similar answers" mean?
You scan or photograph the answer sheets, and it reads the text on its own and brings them in per student. It reads handwriting too, if it's reasonably legible. Up to here it's just input.
The interesting part is what comes next. It doesn't show the hundred and thirty sheets blindly in roll-number order — it groups similar answers and shows them. "Twelve answers that used this concept similarly," that kind of thing. Why does this matter? Because when you see similar answers one after another, you see them by the same standard. When you see them in scattered order, the score you gave student 3 and the score you gave student 47 drift subtly apart. We're human, we waver. You could see the same answer as the first sheet and the hundredth and give it a different score. You get tired, and if the answer right before was exceptionally good, that impression lingers and the next one looks harsher. Grouping similar ones together cuts that wavering sharply. Same-shaped answers, by the same yardstick, in a row. That's something a paper stack simply could never do.
So how do you actually grade on the screen?
When I pick an item, that item's criteria appear as a checklist. As I read the answer, I tap the items that are met, and the score adds up right there on the spot. I jot a one-line note beside it, next answer. At the top, "94/132 sheets graded" counts itself up.
This is the screen I see while grading. Answer on the left, checklist on the right, note at the bottom. Every time I tap an item, the score up top changes on the spot. There's nothing for me to add up. See that black card? "Key concept B missing — 23 students." I didn't count that one by one either; it picked it out on its own. Twenty-three students missing the same concept isn't a sign they failed — it's a signal that I taught it less. One or two students is a student problem; twenty-three is a problem with my lesson. It's not just graded — it shows me what to revisit in the next lesson. This was something you could never see grading on paper. Once you've turned over all hundred and thirty sheets, only the scores remain; where everyone collapsed together doesn't.
What if a student asks, "Why this score?"
This used to be the hardest part. On a paper grading sheet there's only the score and a short note, so I'd have to say "Hmm… this part here…" and peer back into the answer. Now I just show them the items that were checked. "Key concept B is missing here, and your example went off in this part." The student is convinced, and I have a clear conscience.
Grading became explainable. Because the fact that I applied the same standard to everyone in the same shape stays on the screen. It's the same when a parent calls to complain. It doesn't start emotionally with "Why only my child," it starts with "This item was missing." When grading becomes transparent, trust follows. This is something paper, which leaves only a score and a short note, could never do.
But — this tool doesn't grade for you, does it?
Absolutely not. This is the most important distinction. This screen does not read the answers and assign scores in my place. This isn't a story about AI grading essays. Whether this student truly understood the key concept when they wrote it, whether the example fits the context — I still read and judge that. The screen only records my judgment quickly, adds it up, and keeps me from missing anything.
Especially the ambiguous answers, you know? An answer outside the criteria that's oddly good, an answer that has all the concepts but contradicts itself. Those can't be captured by a checklist. Those I stop for, read, write by hand in the note box, and adjust the score manually. Grading that needs interpretation — like literary criticism or deep argumentation — is entirely my job. Where this tool shines is grading that's clear-cut and repetitive. Knowing that boundary is my job too.
"Because the tool takes over the manual work, what's left is judgment. In the same evening, I read more answers, more calmly. It didn't replace me — it widened the space for my judgment."
To sum up, what did you, the teacher, do, and what did makemind do?
What I did was judge. Which items are criteria, what earns full marks and what gets deducted, which answers are tricky, where to stop and read for myself. Only someone who's taught for twenty years knows that. What makemind did was carry that judgment.
| Stage | What I decided (twenty years of judgment) | What the tool did |
|---|---|---|
| Criteria input | Which items belong in grading | Turned items into the grading screen's skeleton |
| Answer input | — | Read text from scans/photos, organized per student |
| Grouped view | What to see by the same standard | Gathered similar answers, reduced the wavering |
| Grading & tallying | Judging whether criteria are met | Auto-summed scores, progress, picked out missed items |
| Screen | What to show and where | Rendered to screen as spoken |
I filled only the left column. The tool took the right. Work that would have taken months if I'd learned to code, or millions of won if I'd outsourced it — I did just by putting my judgment into words.
What changed the most?
My evenings. It's not just that the time shrank. Before, grading was devoured by manual work, so I had no room to actually read the answers deeply. Past midnight, I'd just want to get it over with quickly. That always made me feel guilty. A student spent an hour agonizing over an answer, and I'd flip past it in thirty seconds with a foggy mind.
Now that the manual work moves to the screen, in the same evening I read more answers more calmly. My eyes land properly on each student's writing. Because I never look away from the answer to transcribe a score, my gaze stays on the student's writing the whole time. And when grading is done, things like "twenty-three students missed this concept" become visible, so the next lesson is different. Grading leads into teaching. Before, grading was just a chore to clear away; now it's reading for the next lesson. That's what changed the most.
What would you say to other teachers?
Don't start grand. I didn't build all of this from the start either. I started with a checklist for one item. Tap an item, the score adds up. Once that worked, I added a progress count, added answer input, added grouping of similar answers — I grew it one box at a time.
The key is this. What you know best is your work, not coding. You only have to speak that thing you know well. If the grading criteria already exist in your head, that's already eighty percent of the tool. The moment you lay on the one missing axis, your evenings come back. And those evenings get spent reading more deeply. A screen built by someone like me, who can't write a single line of code, is holding up my grading right now.
makemind.dev "Field" — the person who knows the domain best, building their own tool with their own hands. Han Mi-jeong's grading screen didn't replace her judgment; it widened the space for it.