Glossary

The terms you will meet in Studium's documentation, interfaces and tool results, in alphabetical order. Each entry links to the page that explains it.

Term Meaning More
Agent-drafted lesson A lesson an AI assistant drafted from sources the learner chose, with create_lesson. Not checked by a teacher; its notice says so. Honesty and provenance
Approval record approval.json, written next to a lesson by experience_packet.py approve, bound to the SHA-256 of the exact lesson.json. Documents a teacher's review; not a signature. Honesty and provenance
ASB capsule A packaged, reproducible analysis (from the Agentic Science Builder project) that a reproduction scene asks the harness to run. Scene types
ASTRA, MySTRA A format that records adaptive decisions and analyses so they can be replayed, and its local reader. Used by learner reports and the Observatory. Learner report
Beta mastery The estimate of how well a concept was shown, α / (α + β), fed only by graded answers. Learner model
Bloom level The cognitive level of a concept (remember to create). It sets the difficulty weight of its answers. Scene types
Calibration The gap between the confidence a learner declares and the mastery they have shown. Learner model
Certified mastery A concept at mastery 0.8 or more after at least three graded attempts, with no measured overconfidence or over-trust. A mastery claim is then written. Learner model
Class report class_report: aggregate figures for a course over consented telemetry, each computed over at least five learners. Follow a class
Community bundle A lesson written by a learner and released by a teacher for others. Learner contributions
Concept map The lesson's concepts, prerequisite links and your state on each, with the path travelled and the evidence behind it. Concept map
Contest A claim that you know a reached concept or are unsure about it, checked by one or two questions. The claim alone changes neither mastery nor the path. Concept map
Diagnostic scene An opening scene marked diagnostic: true that asks what the learner already knows. It is played once and never comes back as a review. Review over time
Effective mastery Mastery discounted by recall now: posterior × retrievability, or the posterior alone when never reviewed. Learner model
File channel A task-and-answer loop through files in .studium-session/, for a harness without MCP. Harnesses
FSRS Free Spaced Repetition Scheduler: the model Studium uses to decide when a concept is due for review. Learner model
Guess correction On a choice item, a wrong answer also withdraws the credit a lucky guess would have earned, so random answers do not read as partial mastery. Learner model
Harness The AI assistant or client that drives a Studium lesson: Codex, Claude Code, Copilot, OpenCode, Cursor or the VS Code extension. Harnesses
Help ladder Progressive help after your own attempt: reread, hint, explanation, then the reference with the verdict. A hint discounts the credit of a correct answer. VS Code
Hint discount Each hint shrinks the credit of a correct answer and weakens its review grade. Learner model
Judge A model configured on the server (LITELLM_BASE_URL) that assesses free-text answers. Without one, free text is not assessed. Honesty and provenance
Key class Who holds the key that grades a scene type: ungraded, author_keyed, production or externally_grounded. Scene types
Landmark A line a reproduction's run must return (by default the last non-empty line of its output), checked for presence, never for its value. Scene types
Lane (fast / companion) and surface rank The fast lane is a panel, pane or forms; the companion lane is the conversation. The rank names the surface the session is played on: 3 for a panel or pane, 2 for forms, 1 for conversation. Lesson panels
Learning Workflow Graph (LWG) The graph of concepts and tasks in a lesson, with typed edges such as prerequisite_for. How Studium works
Lesson bundle lesson.json: one lesson's metadata, concepts, graph, scenes and rules. Lesson bundle reference
MCP Model Context Protocol: the standard through which assistants call Studium's tools. MCP tools
Modulator A label (M1, M3, M6) naming the pedagogical intent of an adaptation rule. Modulators
Notice The text a self-study lesson shows first, saying who wrote the keys and what is assessed. Honesty and provenance
Open book A self-study quiz whose key the learner wrote: answering it shows recall of one's own key, not external assessment. Honesty and provenance
Pseudonym The HMAC of a learner id that names their data folder. Privacy and limits
Reference What an answer is compared with after the answer: key points, planted flaws or flawed claims, and a model answer; for a quiz answered wrong, the right choice. Scene types
Retrievability The probability, from FSRS, that the learner still recalls a concept now. Learner model
Review receipt A record of the review items you opened and your verdict on each graded item. A return can be recorded without approving the lesson. Review and approve
Rule 003 The project rule that no content is released to students without a teacher's validation. Honesty and provenance
Run manifest What a reproduction must produce and check, without the expected values. Scene types
Scene One step of a lesson; there are 13 types. Scene types
Self-rating The learner's own rating of recall (again, hard, good, easy). Schedules reviews; never moves mastery. Learner model
Self-study lesson A lesson written for oneself, by the learner or an agent (mode: self_study). Cannot receive teacher approval. Self-study from references
Stability In FSRS, the number of days after which recall falls to the target of 0.9. Learner model
Study plan get_study_plan: reviews due, concepts to relearn, lessons in progress and the next lesson, across all lessons. Review over time
Trust calibration How far a learner over-trusts or over-distrusts a flawed AI reading of a real executed result. Learner model
Universe A variant of a lesson's path: a baseline and a more scaffolded remedial one. How Studium works
Unmeasured No evidence was collected. Studium never reads it as zero, wrong or forgotten. Learner model
Verdict deck A spot_the_flaw form where the learner marks each claim flawed or sound, scored against the key. Scene types
Voice The line that says who wrote or validated the lesson, or who assessed an answer. It distinguishes a teacher, an AI draft and your own key. VS Code
Withheld A class-report figure some learners contributed to, but fewer than the floor of five: hidden, which is neither zero nor unmeasured. Follow a class

View the source of this page on GitHub