Teacher guide
Studium helps you turn sources you trust into a lesson your students can play, with every graded item tied to a passage you can check. You decide what is taught, what the sources allow you to claim, and when a lesson reaches students. A lesson that passes every technical check stays a draft until you approve it; that human decision before any distribution is the project's rule 003.
This page walks through the whole journey and links to the page for each step. To have your
assistant guide you through it, load the studium-teacher skill (see skills).
The journey at a glance
| Step | What you do | Result | Page |
|---|---|---|---|
| Prepare | Choose a question, an audience and the sources you may use | A pilot plan | below |
| Author | Have your assistant draft the lesson from captured passages | A checked draft | Author a lesson |
| Review | Read the plan, the quotes, the keys and the provenance | Your corrections | Review and approve |
| Approve | Record your decision for the exact lesson file | approval.json |
Review and approve |
| Distribute | Give students the lesson folder | Lessons in their library | below |
| Follow a class | Read class-level aggregates, never a single student | A class report | Follow a class |
| Moderate | Review what learners propose to add | Community items | Learner contributions |
Approval also writes a receipt of the items you opened and your verdict on each graded item. To send a lesson back without approving it, record a returned receipt instead; see Review and approve.
The teacher dashboard library: each lesson with who holds its answer keys and its review status.
Before you start
- Install Studium: installation.
- Play a lesson yourself to see what a student receives: first lesson.
- Connect your assistant (Codex, Claude Code, Copilot, OpenCode or another MCP client): harnesses. No extra model provider is needed to author through it.
- Follow the literature to course tutorial once, on the shipped example, before using your own material.
Prepare the activity
Start from a question that asks for more than a computed output, for example "Is a high spectral similarity enough to identify a molecule?". Then separate:
- the concept to understand and the task to carry out;
- the expected observation and the limit to discuss;
- a new case to check transfer.
Decide the pilot: audience, prerequisites, duration, judgement question, what students may explore, which materials are allowed and how results are returned. A framing prompt for your assistant can look like this:
Prepare a Studium draft for [audience], in [duration], on [question]. Use only the passages and materials in the packet. Separate the paper's results, hypotheses and synthetic data. Ask for a prediction before any run, a comparison and a transfer question. Flag missing evidence. Mark nothing as approved and invent no capsule result.
The use cases show complete examples (spectral annotation, diagnosing an analysis, robustness, photographs in archaeology). Ready-made interactive starters are in activity templates; figures, equations and media in media and visual support.
Distribute a lesson
A lesson is a folder: lesson.json, its assets/ and, once approved, approval.json. Keep the
three together and at the same revision. A lesson shipped as a .zip carries the record as its
member approval.json (the demo build scripts add it when it matches lesson.json); without it the
lesson reads "No approval is recorded for this lesson." Students place the folder in their Studium library
(by default ~/.studium/library/<course_id>/, or the folder BUNDLES_ROOT names) and open it
from their assistant or from VS Code (see VS Code). When a learner starts
the lesson, the session reports whether an approval record matches the exact file.
For a class:
- start with a few students and check the session on their machines (bundle access, display, language, behaviour when the model or the network is unavailable, export);
- give written instructions and a fallback if the assistant account or the internet fails;
- the VS Code package makes the lesson pane easy to install, but it does not distribute or validate lessons for you.
Run an experiment session
When a lesson lets students explore with their assistant, set a concrete scope: data, commands, dependencies, web access, interactive controls and duration. Redisplaying the approved materials can stay inside the activity; a new claim, exercise or scene generated beyond that scope is a draft you review before use. Studium does not yet enforce that boundary for you.
Ask students for their prediction, the differences they observed, one limit and a transfer test. An agent's output is not evidence that the student understood. The agentic experiment tutorial gives a full run, and literature and evidence explains the portable exploration receipt.
What students can share with you
A student may choose to export a learner report of their recorded actions, and keeps a private experiment journal. Discuss with them what will be shared and how you will read it: an incomplete trace does not measure autonomy. The default report excludes personal text and gives no access to the full journal or the assistant's history. See privacy and limits.
The interaction report shows how a consenting class used
the interface. To lay those class figures over the lesson's concepts, use tools/concept_map_report.py
and the concept map's teacher viewpoint.
Limits
- Technical checks (schema, exact quotes, lint) are not a pedagogical review; an exact quote can still be misread.
- The approval record names the reviewer you type; Studium does not authenticate it.
- The dashboard, Observatory and analytics services are local tools without login; do not expose them on the internet.