What is a Learning Session?

# Learning Sessions (Learning Units)

Last Updated: 2026-09-04 | For: Teachers & Content Creators | Level: Advanced

What is a Learning Session?

A learning session is a unit where MoFaCTS uses AI-style adaptive scheduling, implemented through configurable cognitive memory models and probability calculations, to select which question to show the student next. Unlike assessment sessions (which have a fixed sequence), learning sessions continuously adapt based on:

• Student's correct and incorrect answers
• Time since last seeing each question
• Overall performance trends
• Spacing and retention intervals

This supports the testing effect by giving learners repeated retrieval practice and concentrating practice on items that the model predicts need more work.

A learning unit may also include unitinstructions; these display before adaptive practice. For a screen with only instructions and no practice items, omit learningsession and use an instruction-only unit instead.

When to Use Learning Sessions

Use learning sessions for:

• Practice and review activities
• Long-term retention training
• Adaptive flashcard-style learning
• Mastery-based learning

Don't use learning sessions for:

• Standardized tests (use assessment sessions instead)
• Fixed-order tutorials
• Pre/post tests requiring consistent ordering

Learning Session Configuration

Prerequisites: understand Stimulus Files, Trial Types Reference, and Learning Algorithms Reference before customizing adaptive behavior.

Resume and Persistence Behavior

Learning sessions now resume from committed history rather than a persisted per-card cache.

• After each completed trial, the canonical saved record is the history row for that trial.
• On resume, MoFaCTS reconstructs learning state from history and recomputes probabilities fresh.
• The next card shown after resume may differ from the card that was on screen before reload if forgetting or other time-sensitive model inputs have changed.
• If a card was displayed but not answered before reload, there is no new committed history for that card, so the unit restarts from its deterministic unit-start state and current history.

Learning sessions are configured in a unit using the learningsession tag:

Learning-session deliverySettings are documented in Content Creation Reference Tables.

Progressive Course Practice

Course Assignments can combine compatible lessons into an ordered progressive lesson. Every member must contain exactly two units: Unit 1 is an instruction unit and Unit 2 is a named learning session. The Unit 2 clusterlist must select valid canonical stimulus clusters, and maxTrials and practiceseconds must be zero or omitted so practice can continue indefinitely.

When a student chooses progressive practice through a particular member, MoFaCTS combines that member with all earlier members. It uses the selected member's Unit 1 instructions and Unit 2 learning model/settings, merges clusters with the same normalized clusterKC, and keeps distinct stimuli separate using the pair stimuliSetId + stimulusKC.

Every saved trial retains its original member TDF, stimulus-set, item, cluster, and Unit 2 identity. As a result, trials completed in the combined runtime reappear when that original lesson is later opened individually. There is no progressive completion state; teachers may change membership or ordering without rewriting or deleting prior learner history.

learningsession is for units that should generate adaptive practice trials. If the intended unit is only a transition, explanation, consent, or completion screen, use an instruction-only unit with instruction fields and no learningsession object.

For first-introduction behavior, deliverySettings.studyFirst can be set to a probability from 0 to 1 to control how often newly introduced items begin with a study trial instead of a normal drill trial. 0 means never, 1 means always, and values such as 0.1, 0.5, or 0.9 give that proportion of first introductions a study trial.

• A calculateProbability formula receives a data-only snapshot. Raw stimulus/response objects, browser/server globals, DOM, Meteor, network APIs, and credentials are never exposed.
• Formulas may use numeric literals, comments, parentheses, semicolons, let/const numeric locals, arithmetic, comparisons, Boolean logic, ternaries, and bounded if/else statements.
• Approved array inputs support numeric indexing, .length, and .slice(). Every helper exposed by the MoFaCTS probability runtime is supported: testFunction, mul, logitdec, recency, quaddiffcor, quaddiffincor, linediffcor, linediffincor, arrSum, errlist, componentSpacing, spacingLagged, ppew, ppet, ppetw, slideppetw, ppes, and ppesFromTimes. All deterministic ECMAScript 2022 Math constants and functions may be freely composed; Math.random remains prohibited.
• A formula may assign p.probability, optional Boolean p.available, and author-named numeric/Boolean/short literal-string diagnostics. Model inputs are read-only.
• Loops, functions, callbacks, recursion, objects/classes, new, imports, async work, computed-property escapes, prototypes/constructors, randomness, and eval/Function are rejected with a field-path and line/column error.

Set p.available = false to make the current stimulus unavailable to the selection algorithm.

Starter calculateProbability model

For authors who want a practical starting point, this model is a reasonable default for many standard practice lessons:

p.y = -0.77 + .665 * pFunc.logitdec( p.overallOutcomeHistory.slice( Math.max(p.overallOutcomeHistory.length-60,  0),   p.overallOutcomeHistory.length),  .966)+ .51* (p.stimSuccessCount) + 11.1 * pFunc.recency(p.stimSecsSinceLastShown,  .443) ; p.probability = 1.0 / (1.0 + Math.exp(-p.y));  return p

This form was adapted from a multiday high school practice experiment supported by NSF and is included here as a useful authoring baseline rather than a universal best model.

Example of Learning session unit:

{
				"unitinstructions": "<p>Welcome to MoFaCTS! </p>\n <p>How it Works: fill-in-the-blank to complete each sentence from your textbook. MoFaCTS helps you practice - you are not graded on your performance, and it expects you to get answers wrong. MoFaCTS automatically moves on to the next flashcard so you can keep practicing at a good pace, and come back to difficult ones later.</p>\n <p>Progress: The Progress page shows how much practice you have done and how well you have learned the material to help you decide if you should keep practicing.</p>\n <p>Audio Settings: You can silently read and type your answers, or the system can read the questions out loud and detect your spoken answers.</p>\n <p>Practice Tips</p>\n <ul>\n <li>Read your textbook before using MoFaCTS</li>\n </ul>\n <ul>\n <li>Use the Refutation and Dialogue Feedback if you need help learning why you got an answer wrong</li>\n <li>This is not a test! It&rsquo;s ok to get answers wrong. The system will help you practice those items.</li>\n <li>If you have any technical problems, use the &ldquo;Report Error&rdquo; button to automatically alert the programmers about your issue.</li>\n </ul>",
				"unitname": "Learning unit",
				"learningsession": {
					"clusterlist": "0-77",
					"unitMode": "distance",
					"calculateProbability": "p.y = -0.77 + .665 * pFunc.logitdec(p.overallOutcomeHistory.slice(Math.max(p.overallOutcomeHistory.length - 60, 0), p.overallOutcomeHistory.length), .966) + .51 * p.stimSuccessCount + 11.1 * pFunc.recency(p.stimSecsSinceLastShown, .443); p.probability = 1 / (1 + Math.exp(-p.y)); return p"
				},
				"deliverySettings": {
					"allowPhoneticMatching": "true",
					"studyFirst": "0.5",
					"drill": "30000",
					"purestudy": "16000",
					"skipstudy": "false",
					"reviewstudy": "6000",
					"correctprompt": "750",
					"fontsize": "24"
				}
			}

unitModes

'distance' using function selectCardAndHintClosestToOptimalProbability

Pick the item closest to optimal probability, where distance is measured in logit units (which solves problems with probability having a cieling).

'thresholdcieling' using function selectCardAndHintBelowOptimalProbability

Pick the item closest to the optimal probability, but below

See Also

• Learning Algorithms Reference - Probability model and unitMode details
• Content Creation Reference Tables - Delivery settings
• Assessment Sessions - Fixed-order tests and quizzes
• Trial Types Reference - Understanding d, t, s, m, n, i, f trial types
• Troubleshooting - Canonical issue triage

Operational Validation

1. Validate at least one real student launch for this unit.
2. Verify intended response/timing behavior in Data Output rows.
3. If media is used, verify canonical media references resolve during runtime.