Cognitive Evidence:
the data that reveals how you learn
GLIA does not ask how you are — it observes how you act. Cognitive evidence is the set of implicit behavioral signals the system collects and interprets to build a user model that updates with every interaction.
Cognitive evidence is the set of observable data generated by a user's behavior during their interaction with a learning system, which the system interprets as signals of the individual's cognitive state, processing profile, and learning trajectory. Unlike explicit assessment — which directly asks the user — cognitive evidence is collected continuously and implicitly through natural behavior.
The distinction between explicit assessment and implicit behavioral evidence is operationally important. Explicit assessment (tests, questionnaires, self-reports) is subject to social desirability biases, does not capture intraindividual variability in real time, and generates additional cognitive load at the moment of collection. Behavioral implicit evidence has its own limitations — it is noisier, harder to interpret — but it is continuous, non-intrusive, and captures the actual state during the task.
Types of cognitive evidence
Response times
Latency between stimulus presentation and user response. Reflects processing speed, perceived difficulty, and cognitive activation state.
Error patterns
Type, frequency, and distribution of errors. Distinguishes between knowledge errors (systematic) and state errors (random on previously solved tasks).
Navigation trajectories
Sequence of movements within the system: what is visited, in what order, for how long, with how many revisits. Reveals processing strategies and friction points.
Review behavior
How frequently the user returns to consult previously presented material. High review may indicate excessive load or weak initial consolidation.
Abandonment patterns
At which points in the sequence the user abandons tasks before completing them. Indicates friction, frustration, fatigue, or ZPD mismatch.
Engagement patterns
Variations in active involvement with content: acceleration, voluntary exploration, elaborated responses. Signals high intrinsic motivation or flow state.
From signal to inference: the interpretation model
Individual normalization. A 4-second response time may be fast for one user and slow for another. The system must calibrate signals against the specific user's baseline, not against a population norm.
Situational contextualization. The same signal can have different meanings in different contexts. A long pause before responding may indicate deep processing (positive) or cognitive blockage (negative) depending on task type and prior session patterns.
Longitudinal integration. A signal appearing once is noise; a pattern repeating under similar conditions is information.
No behavioral evidence interpretation model has absolute certainty. Honest systems acknowledge and manage this uncertainty: they make conservative decisions when inference confidence is low, and adjust the model when predictions are not confirmed.
Cognitive evidence and privacy
Continuous collection of behavioral evidence raises privacy considerations that cannot be ignored. Learning behavioral data is sensitive data: it reveals capacities, difficulties, emotional patterns, and cognitive states. Ethical design requires: transparency about what is collected and how it is used, user control over their own data, minimization of collection to what is strictly necessary, and use of data exclusively for the user's own benefit.
Cognitive evidence in GLIA
GLIA builds each user's cognitive profile exclusively from evidence observed during use. There are no initial diagnostic assessments or standardized tests. The profile emerges from interaction — and belongs to the user.
The system uses cognitive evidence for three purposes: real-time adaptation, refining the longitudinal profile, and detecting trends that require changes in the learning strategy.
The cognitive evidence GLIA collects belongs to the user and is for the user. The system never uses it to classify, compare, or report externally without explicit consent. The goal is a mirror that helps the user understand how they learn — not a surveillance tool.