Learning Analytics for Knowledge Explorer

Complete Behavior Inventory

158 Observable Learner Behaviors

Organized across 18 interaction contexts, every observable action within KE is captured, timestamped, and used as evidence in learner models.

🗺️

Map Generation & Setup

13 behaviors ▼

#	Behavior	Assessment Signal
1–3	Select domain / subdomain / concept	Prior knowledge scope; familiarity with terminology
4	Type custom domain/concept (not from suggestions)	Self-directed learning; specific gap awareness
5	Add additional context before generation	Prior knowledge depth; existing mental model
6–9	Switch to "Describe Problem" mode; analyze & select candidate	Problem awareness; real-world motivation; conceptual discrimination
11	Click "Generate Map"	Full input payload reveals conceptual framing
12	Change LLM provider	Technical sophistication; awareness of AI options
13	Wait duration during generation	Patience; expectation management

🔍

Map Navigation & Exploration

16 behaviors ▼

#	Behavior	Assessment Signal
15	Click concept nodes (blue rectangles)	Conceptual interest; prior knowledge
16	Click skill nodes (green pills)	Application orientation; practical learner
17	Click prerequisite nodes (purple hexagons)	Foundation-seeking; recognizes dependency
18	Click edges (relationship lines)	Relational thinking; seeks connections
19–20	Hover over nodes / edges (without clicking)	Hesitation; consideration depth
21	Drag nodes to reposition	Spatial reasoning; desire for personal organization
22–24	Zoom in/out; pan canvas	Focus: detail-oriented vs. big-picture thinker
25	Change layout algorithm	Visual learning preference; structural curiosity
26–29	Node visit sequence; revisits; navigation paths	Breadth-first = surveying; depth-first = focused; random = disoriented
27	Dwell on map overview before first click	Strategic planning vs. impulsive behavior

📋

Node & Edge Details Panel

5 behaviors ▼

#	Behavior	Assessment Signal
30	Dwell time reading node description	Processing depth; reading speed
31	Dwell time reading edge description	Relational thinking; attention to connections
32	Click "Explore in New Tab"	Curiosity depth; desire to go deeper into periphery
33	View map statistics	Meta-awareness of structure; analytical disposition
34	Time before taking any action	Reflection vs. impulsive exploration

⚡

Learning Actions (Concepts, Skills, Prerequisites, Edges)

14 behaviors ▼

#	Behavior	Assessment Signal
35	View Definition	Foundational learning; beginner orientation
36	Show Prerequisites	Self-assessment; metacognitive awareness
37	See Example	Concretization need; abstract-to-concrete processing
38	Practice Question	Active retrieval; self-testing; strong learning strategy
39	Guided Practice (skill)	Application orientation; hands-on learner
40	Step-by-Step Tutorial (skill)	Procedural learning; scaffolding preference
42	Review Concept (prerequisite)	Gap recognition; builds foundations first
43	Diagnostic Check (prerequisite)	Metacognitive sophistication; verifies understanding
44	Why It Matters (prerequisite)	Meaning-seeking; deep learning orientation
45	Deep Dive (prerequisite)	Mastery pursuit; not satisfied with surface knowledge
46	Explain Relationship (edge)	Relational understanding; seeks "why" not just "what"
47	Show Formal Definition (edge)	Technical sophistication; precision-seeking
48	Common Misconceptions (edge)	Error prevention; error-aware; sophisticated learner

💬

AI Chat Modal

13 behaviors ▼

#	Behavior	Assessment Signal
49	Dwell time before first reply	Processing depth; careful reading vs. skimming
50–51	Compose & send follow-up questions	Comprehension depth; inquiry sophistication
52	"More detail" clicks	Insufficient initial understanding
53	"Simpler" clicks	Cognitive overload; scaffolding needed
54	"Got it" clicks	Self-reported comprehension checkpoint
55	Click suggested follow-up prompt	Guided learning; needs scaffolding for questions
56	"Regenerate" clicks	Quality dissatisfaction; critical evaluation
57	Math input / LaTeX composition	Technical sophistication; formal notation fluency
58–59	Conversation turns; inter-message latency	Engagement depth; cognitive processing time
60–61	Close modal; scroll behavior in chat	Session satisfaction; content revisiting

📝

Assessment Modal

18 behaviors ▼

#	Behavior	Assessment Signal
63–65	Select presets; add custom instructions; skip vs. customize	Self-awareness of level and learning context
67	Answer a question (option selected)	Knowledge state; difficulty calibration
68	Change answer before submitting	Uncertainty; second-guessing; partial knowledge
69	Time per question	Fast+correct = fluent; slow+correct = effortful; fast+wrong = guessing
70	Peek at explanation before answering	Using test as learning tool; help-seeking
71–72	Read explanation after answering; dwell time	Learning from errors; growth mindset
73	Question navigation pattern	Linear = methodical; jumping = strategic or anxious
74	Skip questions	Identifies specific knowledge gaps
75	Score by difficulty tier	Recall (easy) vs. application (medium) vs. synthesis (hard)
77	Assessment retakes	Persistence; score improvement = learning; no change = stuck
79	Abandon assessment	Frustration; overwhelm; disengagement signal

❓

Q2L — Question to Learn Modal

9 behaviors ▼

#	Behavior	Assessment Signal
82	Time reading scenario before first question	Comprehension effort; careful reading
83	Context item selection (which concepts seen as related)	Relational awareness; conceptual mapping ability
84	Question formulation quality	Higher-order thinking; understanding of concept boundaries
85	Number of questions asked	Curiosity persistence; engagement depth
86	AI-evaluated question quality score	Deep vs. shallow understanding
87	Own questions vs. suggested prompts ratio	Inquiry independence; autonomous thinking

🔧

WTB — When Things Break Down

13 behaviors ▼

#	Behavior	Assessment Signal
129	Open WTB modal	Engagement with troubleshooting mode; causal thinking orientation
130–131	Select presets (Engineering/Biological/Social/etc.) vs. skip preferences	Domain preference; self-awareness of learning context
132	Dwell time reading breakdown scenario before first action	Comprehension effort; careful reading vs. rushing to diagnose
133	Expand evidence clues (type + order + total expanded)	Evidence-gathering strategy; systematic vs. selective reading
134	Ask diagnostic investigation questions (turn count, clues read at time of asking)	Investigative depth; ability to formulate targeted diagnostic queries
135	Switch to diagnosis mode (investigation turns + clues expanded at switch)	Readiness calibration; how much evidence gathered before committing
136–137	Compose & submit root cause diagnosis (text length, composition time, `wentDirectToDiagnose`)	Causal reasoning quality; overconfidence if direct without investigation
138	Diagnosis accuracy verdict (correct / partial / incorrect + reasoning depth)	Direct measure of Apply/Analyze Bloom's levels; mechanism vs. symptom identification
139	Retry diagnosis vs. continue to discussion	Metacognitive calibration; persistence after failure
140–141	Post-verdict discussion turns; close WTB (session duration, phase at close)	Post-diagnosis conceptual integration; depth of follow-through

WTB Causal Reasoning Signals: The combination of clues expanded before diagnosis (evidence coverage), investigation turns taken (diagnostic thoroughness), and diagnosis accuracy (correctness) forms a multi-dimensional measure of causal reasoning ability — one of the hardest cognitive skills to assess through traditional tests.

🔭

Wonderment — Curiosity to Insight

17 behaviors ▼

#	Behavior	Assessment Signal
142	Open Wonderment modal	Orientation toward inquiry-based learning
143–146	Select inquiry lens presets; add custom instructions; toggle fresh generation	Metacognitive awareness of own learning preferences; preference for novelty
147–148	View observations list; dwell before selecting	Reflective reading — long dwell = careful consideration of which curiosity resonates
149	Write own custom observation (`compositionMs`)	Independence signal — self-directed inquiry; has existing curiosity the AI didn't surface
150	View academic reframing of observation	Exposure to disciplinary framing; field label provides vocabulary scaffolding
151	Select genius question (`type` + `geniusDwellMs`)	Question type reveals thinking preference; long dwell = deliberate; "paradox" and "inversion" types correlate with higher-order thinking
152	Navigate back to genius questions (with draft)	Metacognitive self-correction — critical self-assessment; intellectual honesty
153	Compose and submit answer (`compositionMs`)	Long composition = genuine engagement; short = low effort or overconfidence
154–155	Read critics' panel (`criticsDwellMs`); request synthesis	Long `criticsDwellMs` = open to critique; rushing = seeking validation not learning
156	Read synthesis and wonderment moment (`synthesisDwellMs`)	Conceptual integration signal — longest dwell indicates concept has been genuinely reframed
157–158	Restart / close (`phase` at close, `didReachSynthesis`)	Drop-off phase identifies most difficult or disengaging step; completion = full inquiry arc traversed

Bloom's Levels 5–6 Direct Measurement: Wonderment is the only activity that operationalizes Evaluate (critics' panel engagement, synthesis dwell) and Create (custom observation, genius question selection, answer composition) simultaneously — the two Bloom's levels hardest to assess through traditional tests. The critics' dwell vs. synthesis dwell ratio distinguishes learners who seek validation from those who genuinely integrate new knowledge.

🎓

Lecture, Resources, & Learning Path

26 behaviors ▼

#	Behavior	Assessment Signal
92	Slide dwell time	Content processing; long on complex slides = careful reading
93	Lecture slide navigation (linear vs. backtracking)	Didn't understand on first pass = re-processing need
94	Complete lecture (all slides)	Commitment to structured content
96	Abandon lecture midway	Content mismatch; boredom; frustration
99–100	Resource type filter; click external link	Learning modality: video=visual, article=textual, tutorial=hands-on
103–104	"Suggest Learning Order"; "Where Should I Start?"	Orientation need; self-regulation level
105	Follow vs. deviate from suggested path	Compliance vs. self-directed exploration
107	Mark nodes complete	Self-monitoring; progress tracking disposition
108	Complete out of order (before prerequisites)	Knowledge overconfidence; skips foundations

🗂️

Tab Management, Session & Dashboard

20 behaviors ▼

#	Behavior	Assessment Signal
109–112	Open/switch/close tabs; concurrent tab count	Comparative learning; simultaneous concept exploration
115	Total session duration	Overall engagement depth
116	Return visit patterns	Commitment; sustained motivation; spaced repetition
117–118	Dashboard visits; expand domains	Metacognitive reflection; progress awareness
121	"Learn Again" re-engagement	Spaced repetition awareness; optimal strategy
122	View "How Am I Doing?" analytics	Self-monitoring; metacognitive engagement
125–128	Class enrollment; curriculum adherence; pace	Social context; structured vs. self-directed learning

Computational Models

8 Learner Models Derived from Behavior

Raw behavioral data feeds into a stack of learner models, each measuring a different dimension of the learner's profile. Together they form a complete, dynamic picture of the learner.

Model 1

Knowledge State Model

Bayesian Knowledge Tracing (BKT)

Estimates probability that the learner has mastered each concept/skill at any moment. Uses assessment scores, "Got it" / "Simpler" feedback, revisit patterns, and Q2L quality as evidence to update P(Known) per concept.

P(Known) per concept — 0.0 to 1.0

Concept map propagation via edge structure

Fast+correct = strong signal; slow+wrong = guessing

Model 2

Learning Strategy Profile

Latent Class / Gaussian Mixture Model

Classifies each learner into a strategy archetype based on a 13-dimensional feature vector derived from time allocation, exploration entropy, question formulation, and interaction patterns.

Primary archetype classification (soft membership)

Preferred modality: visual / textual / hands-on

Personalized recommendations per archetype

Model 3

Engagement & Motivation Model

Hidden Markov Model (HMM)

Tracks motivational state over time through 5 hidden states — from Highly Engaged to Disengaging — using session duration, question depth, completion rates, and return frequency as observable emissions.

Current engagement state (S1–S5)

Transition probabilities between states

Alert when P(Disengaging) exceeds threshold

Model 4

Bloom's Taxonomy Level Estimator

Evidence Accumulation Model

Estimates the highest cognitive level demonstrated per concept, from Remember (L1) through Create (L6), using behavioral evidence: definitions viewed, examples requested, chat question quality, misconceptions explored, and Q2L originality.

Bloom level (1–6) per concept

Evidence count per level threshold

Cognitive depth progression over time

Model 5

Concept Mastery Map

Weighted Overlay Model

A composite mastery score (0–100) per node, combining assessment performance (30%), Bloom level (20%), chat engagement depth (15%), time invested (10%), spaced repetition bonus (10%), prerequisite mastery (10%), and resource diversity (5%).

Visual color overlay on the concept map

Propagation rules via requires/enables edges

Mastery decay (exponential, 30-day half-life)

Model 6

AI Collaboration Competency

Multi-Dimensional Scoring

Evaluates how effectively the learner uses AI as a cognitive partner across 6 dimensions: prompt precision, iteration depth, critical evaluation, concept integration, metacognitive awareness, and efficiency.

Score per dimension (0–100)

Overall AI collaboration competency score

Independence trajectory over sessions

Model 7

Temporal Learning Trajectory

Growth Model / Time Series

Tracks learning velocity, breadth expansion, depth progression, and efficiency trends over time. Predicts which concepts will need review, estimates time-to-mastery, and flags learners who are plateauing.

Learning velocity (Δmastery / Δtime)

Forgetting curve predictions per concept

Strategy maturation: passive → active learning shift

Model 8

Social-Comparative Model

Percentile Ranking in Class Context

Positions learner performance relative to class peers across curriculum progress, assessment scores, engagement consistency, exploration breadth, and AI collaboration quality. Enables teacher interventions and peer grouping.

Percentile rank per metric within class

Class-wide concept mastery heat map

At-risk learner identification and pacing alerts

Assessment Signals

What Each Behavior Reveals

Each interaction is tagged with the cognitive or non-cognitive characteristic it indicates. Together they paint a complete learner portrait — without a single explicit test.

Bloom's Taxonomy Progression

6
Create
Describes novel problems · Cross-domain connections · Original Q2L questions · Multi-tab synthesis · Wonderment: custom observation + genius question type "reframe/inversion"
5
Evaluate
High Q2L scores · Critical inquiry · Challenges AI output (Regenerate + critique) · Multiple perspectives · Wonderment: long critics dwell + synthesis integration
4
Analyze
Edge exploration · "Why" & "How" questions · Common Misconceptions · Diagnostic Check · WTB: correct root-cause diagnosis with deep reasoning
3
Apply
Guided Practice · Step-by-Step Tutorial · Application questions in chat · Medium questions correct · WTB: diagnostic investigation questions asked
2
Understand
Asked for examples · "More detail" clicks · Time reading explanations · Medium questions correct
1
Remember
Viewed definition · Correct on easy assessment questions

WTB — Causal Reasoning Signals

Cognitive

Clue coverage before diagnosis

Reads all clues = systematic evidence synthesis; skips clues = impulsive or overconfident

Cognitive

Investigation turns before committing

More questions = thorough diagnostic process; zero questions + wrong = knowledge gap or guessing

Cognitive

Diagnosis accuracy (correct / partial / incorrect)

Direct measure of Apply/Analyze Bloom's levels; most reliable signal in the system

Cognitive

Reasoning depth: mechanism vs. symptom

Identifies WHY it broke (deep) vs. WHAT broke (shallow) — distinguishes conceptual understanding from surface recognition

Metacognitive

Went directly to diagnose without investigating

If correct = strong prior knowledge; if incorrect = overconfidence; triggers adaptive scaffolding

Metacognitive

Retry vs. continue after wrong verdict

Retry = growth mindset + self-regulation; continue despite being wrong = needs metacognitive intervention

Behavioral

Scenario preset selection

Domain preference and contextual learning orientation (engineering, biological, social, etc.)

Behavioral

Post-verdict discussion depth

Follow-up questions about mechanism or prevention signal genuine conceptual integration beyond the task

Wonderment — Inquiry Depth Signals

Cognitive

Custom vs. selected observation

Writing own = active prior curiosity; selecting AI's = receptive but less self-directed — strongest independence signal in the system

Cognitive

Genius question type chosen

"Paradox" and "inversion" types correlate with higher-order thinking; "scale" and "analogy" indicate systems thinking across domains

Cognitive

Genius question dwell time

Long consideration before selecting = deliberate, careful thinker; very short = impulsive or already had a strong prior preference

Metacognitive

Back to questions (with draft)

Self-correction behavior — had an answer started but chose to reconsider the question; high metacognitive awareness

Metacognitive

Critics panel dwell vs. synthesis dwell ratio

High critics dwell + high synthesis dwell = genuine integration; low critics + high synthesis = seeking validation; low both = surface engagement

Cognitive

Synthesis dwell time

Strongest single indicator of deep conceptual impact — when the wonderment moment genuinely reframes understanding, learners read slowly and return

Behavioral

Phase at close (drop-off analysis)

Identifies which step is too difficult, too abstract, or least engaging — directly informs adaptive content and pacing decisions

Affective

Answer composition time

Long composition = high engagement and genuine effort; very short = low investment or overconfidence in the question chosen

Learning Strategy Archetypes

📐

Systematic Explorer

Follows prerequisite→concept→skill order · Reads definitions first · Uses learning path · Linear assessment navigation

→ Provide structured curricula; suggest next logical concept

🌐

Curiosity-Driven Wanderer

Opens many tabs · Jumps between non-adjacent nodes · Explores edges · Many follow-up questions

→ Provide open-ended exploration tools; suggest surprising connections

🏆

Assessment-Focused Achiever

Goes straight to assessments · Retakes for higher scores · Short chat sessions · Skips lectures

→ Gamify with score tracking; challenge with harder questions

📖

Passive Consumer

Primarily uses lectures · Reads definitions · Few follow-up questions · Rarely takes assessments

→ Prompt with practice questions; interleave active recall

🆘

Help-Seeking Struggler

Frequent "Simpler" clicks · Long time per question · Revisits prerequisites · Uses "Where Should I Start?"

→ Scaffold more; recommend prerequisite review first

⚡

Efficient Expert

Short focused sessions · High assessment scores · Explores misconceptions · Uses formal definitions

→ Provide advanced content; suggest edge cases; cross-domain connections

Engagement State Model (HMM)

S1 — Highly
Engaged

Deep Exploration

Long sessions · High question count · Explores all node types · Completes assessments · Moderate "Got it" rate · 1–2 days between sessions · WTB: systematic clue reading + correct diagnosis

S2 — Productively
Engaged

Focused Progress

Medium sessions · Regular interactions · Steady pace · Frequent "Got it" clicks · Dashboard visits · 2–3 days between sessions

S3 — Surface
Engaged

Going Through Motions

Short sessions · Few questions · Random node exploration · Partial assessment completion · Rare dashboard visits · 3–5 days between sessions

S4 — Struggling

Friction & Confusion

Long pauses · Prerequisite revisits · Frequent "Simpler" clicks · Abandoned assessments · Frequent self-check via dashboard

S5 — Disengaging

⚠️ Intervention Needed

Very short sessions · Minimal node visits · Declining return frequency · Skips assessments · No follow-up questions · 7+ days between sessions

Concept Mastery Score — Composite Formula

            mastery(node) =

              0.30 × assessment_score

             + 0.20 × bloom_level / 6

             + 0.15 × chat_engagement_depth

             + 0.10 × time_invested [diminishing returns]

             + 0.10 × revisit_pattern [spaced repetition bonus]

             + 0.10 × prerequisite_mastery

             + 0.05 × resource_diversity

0–30: Unexplored / Failing

31–50: Partial

51–70: Developing

71–85: Proficient

86–100: Mastered

Model	Data Captured	Assessment Value	Effort	Priority
Knowledge State (BKT)	70%	Very High	Medium	P0
Concept Mastery Map	80%	Very High	Low	P0
Bloom's Taxonomy Estimator	60%	High	Medium	P1
AI Collaboration Model	90%	High	Low (enhance)	P1
Learning Strategy Profile	50%	High	Medium	P1
Engagement / Motivation HMM	40%	Medium-High	High	P2
Temporal Learning Trajectory	60%	Medium-High	Medium	P2
Social-Comparative Model	70%	Medium	Medium	P2

Knowledge Explorer as a
Stealth Assessment Instrument

What is Stealth Assessment?

The Core Idea

158 Observable Learner Behaviors

8 Learner Models Derived from Behavior

What Each Behavior Reveals

Bloom's Taxonomy Progression

WTB — Causal Reasoning Signals

Wonderment — Inquiry Depth Signals

Learning Strategy Archetypes

Engagement State Model (HMM)

Concept Mastery Score — Composite Formula

The Unified Learner Profile

Model Implementation Priority

Knowledge Explorer as a Stealth Assessment Instrument

What is Stealth Assessment?

The Core Idea

158 Observable Learner Behaviors

8 Learner Models Derived from Behavior

What Each Behavior Reveals

Bloom's Taxonomy Progression

WTB — Causal Reasoning Signals

Wonderment — Inquiry Depth Signals

Learning Strategy Archetypes

Engagement State Model (HMM)

Concept Mastery Score — Composite Formula

The Unified Learner Profile

Model Implementation Priority

Knowledge Explorer as a
Stealth Assessment Instrument