Teacher view · class patterns, not individuals

What 28 students learned together. Not what any one of them typed.

You see the shape of the class — where they got stuck, where curiosity clustered, where they asked to talk to a human. Individual transcripts stay between the student and the system. Your job stays teaching, not surveillance.

No per-student logs. You cannot open one student's thread. No comparisons. No leaderboards, no ranks, no individual scores. You see asks. When students tag "needs a teacher", you find out.

Viewing as Teacher · Ms. Liang · AP US History — period 3

28 enrolled · 19 students used the system this week · last sync 12 min ago

Visibility scope: class aggregate only

students active this week

students tagged "ask teacher"

confusion clusters detected

primary sources opened

Where the class is stuck

Topics where 3 or more students arrived at the same confusion within 7 days. Aggregated, never linked to individuals.

"Reconstruction was a failure, but Reconstruction Amendments worked — which is true?"

7 students · this week · several reached low-confidence resolutions on different sides

cluster · 7

Causes of WWI — students unable to distinguish "trigger" vs "underlying cause"

5 students · same week we covered the topic in class

cluster · 5

"Why does the textbook say the Industrial Revolution started in 1760?"

3 students · all resolved high-confidence after reading primary sources

resolved · 3

Students who asked to talk to you

Threads they explicitly paused with a "needs a teacher" tag. You see the topic and the date. You don't see the thread.

Reconstruction Amendments topic

2 students · tagged within last 2 days

awaiting you

Essay structure for the Civil War question

1 student · tagged 4 days ago

awaiting you

"How do I push back on a primary source author?"

1 student · this morning

awaiting you

What this tells you about your teaching

Signals from the week. The system surfaces these for reflection — not for evaluation.

7 students independently chose to dig into Reconstruction beyond what the textbook covered

Suggests engagement with the topic. Their conclusions diverged — worth a class discussion next session.

engagement signal

The "trigger vs underlying cause" distinction landed only for ~40% of students who attempted it

This was a key concept in last week's lecture. Worth re-anchoring with a concrete WWI exercise.

re-teach signal

11 students opened primary sources without being told to

The curiosity habit is forming. Whatever you did to seed this — keep doing it.

strength signal

Metrics this dashboard refuses to show you

These are not "coming soon". They are intentionally not in the product. Each one would change what the class becomes.

Per-student usage time

Encodes "time spent" as the proxy for learning. The student who thinks 10 minutes and writes a great answer is not behind the student who fidgets for an hour.

Individual student transcripts

If students know any AI conversation might be read by you, they stop using AI to think and start using it to perform for you. The middle of learning has to be private.

Leaderboards / rankings

The product is not a comparison engine. AI gives every student a private tutor; ranking them on top of that re-creates the dynamic the tutor was supposed to fix.

"Suspected cheating" alerts

There is no version of this that doesn't punish honest exploration. If the model can't tell, neither can we — and the cost of false positives is too high.

Cross-class comparisons of your teaching

Your colleagues' classes are not shown next to yours. Teaching is not a tournament. Surfacing this would push toward gaming the system rather than teaching.

What stays sharp when AI is in the classroom

AI in education has a default failure mode: it replaces the relationship between teacher and student with a metric pipeline between teacher and dashboard. These six principles push back.

You see clusters, never individuals

If 5 students hit the same confusion, you see "5 students". You do not see who. You decide what to do — re-teach, set up office hours, write a worksheet.

What this prevents — AI as a surveillance layer for individual student behaviour. The teacher-student trust stays intact.

"Ask teacher" tags are the only individual signal

When a student explicitly tags a thread "I want to talk to my teacher about this", the system tells you. The topic; not the conversation. That tag is the student inviting you in — not the system reporting on them.

What this prevents — AI deciding when a student "needs intervention". The student decides.

Engagement signals point to your teaching, not their effort

"11 students opened primary sources" is information about whether your curriculum and your framing seeded that behaviour. Read it as data about your work, not theirs.

What this prevents — student effort scoring. There is no "engagement score" attached to individual names.

You cannot prompt the AI to evaluate students

There is no "summarise how my class is performing" button that wraps individual data in aggregate language. The aggregation happens at the data layer, not the prompt layer.

What this prevents — clever prompt-engineering to recover individual data the UI hides. The privacy boundary is structural.

Re-teach signals are gentle, not graded

"40% landing rate on trigger-vs-underlying cause" is offered as a reflection cue, not a performance review. Your evaluation as a teacher is not happening on this dashboard.

What this prevents — administrators turning these metrics into teacher KPIs. The dashboard surfaces signals to you, not to your principal.

You can ask the class. You cannot ask the system about the class.

If you want to know what students are thinking, the answer is to ask them. The dashboard hints at where to ask, what to ask about, and which students opted in. The conversation still happens human-to-human.

What this prevents — AI mediating the teacher-student relationship. The human moments stay human.

See how students learn → See the parent view Why we built it this way