Concepts

This guide explains the concepts used in my research and teaching. Some are established ideas from the literature; others are labels used in a particular study or teaching routine. Each page identifies that scope and links to its sources.

Browse the concepts

Designing and testing assignments

Iterative perturbation process — Iterative assignment perturbation is an instructor-led process of revising a question and testing a target chatbot’s responses against a rubric.

One-Shot GenAI Resistance — In this research, One-Shot GenAI Resistance means that a specified chatbot scores below the study’s rubric threshold on three attempts with the same question.

Three-Attempt Consistency Protocol — The Three-Attempt Consistency Protocol repeats the same question in three separate chatbot attempts before applying the study’s resistance criterion.

Spatial injection — Spatial injection adds a visual or geometric constraint to a question as part of assignment perturbation.

Axiomatic replacement — Axiomatic replacement changes or explicitly restates the premises and rules a solution is allowed to use.

Contextual embedding — Contextual embedding places a question in a specified setting and requires the solution to use that setting’s definitions or constraints.

Examining student interactions

DRIVE framework — DRIVE is a framework and taxonomy for examining how students direct a generative AI interaction and express their expertise within it.

Cognitive scaffolding — In the AIED case study, cognitive scaffolding names an interaction pattern in which a student recognized an incorrect chatbot response and supplied context or criticism to improve it.

GenAI dependence — In the AIED case study, GenAI dependence names an engagement pattern involving acceptance of incorrect chatbot output without critical evaluation visible in the interaction.

Online, on-demand CS coursework — Online, on-demand CS coursework is computing instruction learners access online and work through on their own schedule.

Teaching and checking reasoning

AI literacy — AI literacy concerns the competencies people use to evaluate AI, work with it, and use it in their lives.

Hallucination autopsy procedure — The hallucination autopsy procedure is a teaching routine for checking a chatbot’s proof, locating an error, and examining a possible explanation for it.

Axiom reversion — Axiom reversion is the teaching label used here when a response substitutes a familiar theorem or formula for the one specified in the problem.

Invalid syntactic shortcut — An invalid syntactic shortcut is a proof step that looks like a familiar manipulation but is not licensed by the relevant rule or assumptions.

Spatial or counting miscount — A spatial or counting miscount occurs when a solution includes forbidden configurations, misses valid ones, or counts a configuration more than once.

Strong prompt structure — Strong prompt structure is a four-part teaching routine for stating context, requesting justification, examining assumptions, and planning corrections.

Collaborative prompting — Collaborative prompting is a teaching routine for working through a problem with a chatbot while keeping responsibility for checking and explaining the reasoning.

Subgoal label — A subgoal label names the purpose of a group of solution steps so a learner can recognize that purpose in another problem.

Capabilities and development practice

Free-tier accessible model — A free-tier accessible model is a model available through an access route that does not require a paid subscription at the time of use.

Structured unevenness — Structured unevenness is the descriptive label used here for a pattern of model performance that varies systematically across subject areas.

Calibrated trust — Calibrated trust is trust that is appropriately related to evidence about a system’s capabilities and limitations in a particular task.

GenAI-assisted prototyping procedure — The GenAI-assisted prototyping procedure is a development workflow that combines a coding agent’s output with human review, testing, play, and handoff.

Connections among the concepts

The map records selected relationships described in the linked papers and teaching materials. An arrow means the relationship stated beside it; it does not establish a learning prerequisite, a causal effect, or a validated model of student knowledge.

FitFull sizeLarger

From	To	Kind	Relationship	Source
Iterative perturbation process	One-Shot GenAI Resistance	prerequisite	The assignment-design procedure produces a question that meets the study's resistance criterion.	Table 1
Iterative perturbation process	Three-Attempt Consistency Protocol	prerequisite	The procedure's last step is where the three-attempt protocol runs.	Table 1, step 8
Three-Attempt Consistency Protocol	One-Shot GenAI Resistance	prerequisite	A question meets the resistance criterion only after the three-attempt protocol has scored it below the threshold on three separate attempts.	Table 1, step 8
Iterative perturbation process	Spatial injection	strategy	The procedure selects spatial injection when a question's reasoning is purely symbolic.	Table 1, step 5
Iterative perturbation process	Axiomatic replacement	strategy	The procedure selects axiomatic replacement when a question can be answered by reciting a standard proof.	Table 1, step 5
Iterative perturbation process	Contextual embedding	strategy	The procedure selects contextual embedding when a question is stated in general terms.	Table 1, step 5
Spatial injection	Axiomatic replacement	relational	Spatial injection changes what a question asks the reader to picture; axiomatic replacement changes what it allows them to assume. Step 5 chooses between them.	Table 1, step 5
Axiomatic replacement	Contextual embedding	relational	Axiomatic replacement changes what a question allows a solution to assume; contextual embedding changes the setting it has to be answered in. Step 5 chooses between them.	Table 1, step 5
One-Shot GenAI Resistance	Cognitive scaffolding	integrative	A question that meets the resistance criterion puts an incorrect response in front of a student, which is what makes cognitive scaffolding observable in a transcript.	Section 3
One-Shot GenAI Resistance	GenAI dependence	integrative	The same question makes GenAI dependence observable: a student who accepts the incorrect response submits different work from one who challenges it.	Section 3
DRIVE framework	Cognitive scaffolding	strategy	The AIED study coded transcripts with DRIVE, which is how cognitive scaffolding was separated out as a pattern.	Section 3.2
DRIVE framework	GenAI dependence	strategy	The same coding separated GenAI dependence out as the other pattern.	Section 3.2
Cognitive scaffolding	GenAI dependence	relational	The analysis reported these as two distinct patterns within the same small sample, rather than as two ends of one scale.	Section 3.2
Online, on-demand CS coursework	GenAI dependence	relational	Participants reported turning to LLMs instead of online coursework, which is the behaviour the later study looked for inside transcripts.	Abstract
Hallucination autopsy procedure	Axiom reversion	strategy	Step A5 of the autopsy procedure is where a reader decides whether the error in front of them is axiom reversion.	Procedure A, step A5
Hallucination autopsy procedure	Invalid syntactic shortcut	strategy	Step A5 is where a reader decides whether the error is an invalid syntactic shortcut.	Procedure A, step A5
Hallucination autopsy procedure	Spatial or counting miscount	strategy	Step A5 is where a reader decides whether the error is a spatial or counting miscount.	Procedure A, step A5
Axiom reversion	Invalid syntactic shortcut	relational	Both produce a proof that reads correctly. They are told apart by which check you run: the closed form against the problem, or the step against the rule it claims.	Slide 16
Invalid syntactic shortcut	Spatial or counting miscount	relational	An invalid syntactic shortcut fails a symbolic check and a miscount fails an arithmetic one, so the two are separated by which check you run.	Slide 16
Axiomatic replacement	Axiom reversion	integrative	After a question's axioms are replaced, axiom reversion is the error to look for: the response may use the result the question removed. It is not an outcome the strategy guarantees.	Slide 16, against Table 1 of the AIED paper
Spatial injection	Spatial or counting miscount	integrative	After a spatial constraint is added, a spatial or counting miscount is the error to look for, rather than an outcome the strategy guarantees.	Slide 16, against Table 1 of the AIED paper
One-Shot GenAI Resistance	Hallucination autopsy procedure	integrative	A question that meets the resistance criterion gives the autopsy procedure something to work on, because the response is wrong often enough to be worth examining.	Slides 10 and 17
Cognitive scaffolding	Hallucination autopsy procedure	integrative	The autopsy procedure writes down, as something that can be taught, the checking behaviour some students already showed.	Slides 16 to 18
GenAI dependence	Collaborative prompting	strategy	Collaborative prompting is the routine taught in response to the dependence pattern: the learner keeps responsibility for checking and explaining the reasoning.	Procedure C, step C1
Strong prompt structure	Collaborative prompting	prerequisite	A prompt written with the four-part structure is what collaborative prompting continues, so the structure comes first.	Procedures B and C
Strong prompt structure	GenAI-assisted prototyping procedure	integrative	The brief, plan and reconnaissance stages each name the prompt-structure step they are carrying out.	Slide 53
Collaborative prompting	GenAI-assisted prototyping procedure	integrative	The stages where a person reads the plan, plays the build and deploys it apply collaborative prompting to software rather than to a proof.	Slide 53
Hallucination autopsy procedure	GenAI-assisted prototyping procedure	integrative	The notebook stage applies the autopsy to an agent's own work: name the rule it used, find the first divergence, classify it.	Slides 34 to 36
Subgoal label	Hallucination autopsy procedure	prerequisite	Steps A0 to A6 are subgoal labels, which is what makes the autopsy a labelled procedure rather than a checklist.	Procedure A, slide 17
Subgoal label	Strong prompt structure	prerequisite	The four parts B1 to B4 are subgoal labels, each naming what that part of a prompt is for rather than what to type.	Procedure B, slide 19
Subgoal label	Collaborative prompting	prerequisite	The steps C1 to C6 are subgoal labels, and the same label appears in the reference box, the worked example and the student's own task.	Procedure C, slide 21
AI literacy	Collaborative prompting	strategy	Questioning the system that produced an answer is part of the critical-evaluation competency, and it is the last step of collaborative prompting.	Slide 52
Free-tier accessible model	Structured unevenness	prerequisite	Subject-level unevenness is measured on the models a student can open without a paid subscription.	Methods
Structured unevenness	Calibrated trust	integrative	Evidence that performance varies by subject is what a learner would use to calibrate trust subject by subject rather than evenly.	Scholarly Significance
AI literacy	Calibrated trust	prerequisite	The strengths-and-weaknesses competency asks a learner to hold an accurate picture of what a tool handles reliably, which is what calibrated trust describes.	Theoretical Framework
Calibrated trust	GenAI dependence	relational	A learner who trusts a tool evenly cannot tell the tool's weak subject from their own, and that confusion is what can leave a wrong answer unchallenged.	Theoretical Framework

For the formal assignment-design procedure, start with the AIED paper. For the teaching routines, see the classroom-study preprint and course materials.