Sparkurity

Methodology

Our approach

Sparkurity is designed on findings from meta-analyses, not single studies. Most of that evidence comes from school and university learners rather than employees, so we say the platform is designed on these findings — never that it guarantees their effect sizes.

The learning graph

How one learner moves through a course.

Five steps and one loop. Every step is built on a research finding — follow the tags to the evidence below. Where the AI suggests, the instructor decides.

  1. 1

    Profile

    Field of work, role, experience and goals. Required before enrolment.

    Relevance ↓
  2. 2

    Skill check

    A diagnostic quiz across the whole course produces a knowledge note per module.

    Instructor approves the quiz

    Expertise reversal ↓
  3. 3

    Personal roadmap

    Each module is marked with one of four actions.

    • Skip
    • Review carefully
    • Standard pace
    • Focus area

    Instructor approves every skip

    Adaptive tutoring ↓
  4. 4

    Adaptive modules

    Retrieval checks inside lessons. Exercises fade from support to independence.

    1. Worked example
    2. Completion task
    3. Independent task
    Worked-example fading ↓
  5. 5

    Mastery check

    Gate quizzes and rubric-graded written answers before the next phase unlocks.

    ✓ Next phase unlocks↺ or loop back

    Instructor reviews AI grading

    Mastery learning ↓
↺ Not yet mastered

Back to step 3: the roadmap regenerates after a failed gate quiz, or when the instructor flags a submission. Feedback ↓

Cohort layer

In cohort courses, the same path runs alongside the group.

  • Role-play scenarios
  • Cohort discussion
  • Live sessions
Collaboration ↓

Cohort usage

Where the cohort fits in a week of learning.

Self-study is individual and adaptive. Live time is for what only a group and an expert can give. Example — week 2 of the Internal Cybersecurity Auditor cohort.

Before the sessionLive session (2 h)After the session
Youindividual, adaptiveSelf-study on ISO 27001 controls and evidence sampling. Skip what your skill check shows you know.Play auditor or auditee in an interview role-play, using your role's briefing pack.Collect and evaluate evidence for the cohort's simulated audit.
Your cohort4–20 peersDiscussion prompt in the cohort channel: what evidence would convince you?Roles rotate. Each role sees only what that role would really have.Compare findings in the discussion thread; challenge each other's evidence.
Instructorin chargeApproves roadmaps; the progress heatmap shows who is stuck.Runs the session; stops the role-play to critique in the moment.Reviews every written assignment; overrides or regrades AI drafts.
AIsuggests, never decidesTutor answers from the course material and standards, with citations.Live time stays human.Drafts rubric-based feedback for the instructor to review.
Live contact hours — Internal Cybersecurity Auditor
Classroom version 32 h mostly theory
Sparkurity cohort 8 h cases, role-play, critique — plus adaptive self-study

Fewer hours in the classroom, more time practising. Theory moves into self-study, where experienced participants skip what they already know.

An adaptive path: start where you are

  • Personalised, step-level guidance works. Intelligent tutoring systems showed a median +0.66 SD over conventional teaching across 50 evaluations (Kulik & Fletcher, 2016) and +0.41 across 107 effects (Ma et al., 2014). Step-based tutors come close to human tutors (VanLehn, 2011).
  • Experts shouldn’t sit through novice content. Support that helps novices slows experts down — the expertise reversal effect (Kalyuga et al., 2003). In one randomised trial, adaptive skipping raised learning efficiency with no loss in scores (Kerfoot, 2010).
  • Difficulty grows with competence. Worked examples fade into independent work as learners improve (Kalyuga, 2007).
  • Mastery before moving on. Mastery learning raised achievement across 108 studies, more for learners who start behind (Kulik, Kulik & Bangert-Drowns, 1990).
  • Testing yourself beats re-reading. Retrieval practice shows g = 0.50, and 0.73 with corrective feedback (Rowland, 2014).
  • Feedback that explains, not just a score. Highly informative feedback is far more effective than a grade alone (Wisniewski et al., 2020) — and poorly designed feedback can hurt (Kluger & DeNisi, 1996), which is why every AI draft is reviewed.
  • Cases that fit your job. Relevance drives motivation and achievement (Göksu & Bolat, 2021); problem-based learning improves skills (Dochy et al., 2003).

Cohorts: learn with people like you

  • Learning together beats learning alone. Small-group learning improved achievement and persistence (Springer et al., 1999); computer-supported collaboration improved skills (Chen et al., 2018).
  • Structured roles make collaboration work. Collaboration scripts with roles strongly improve collaboration skills (Vogel et al., 2017).
  • Interaction is what makes online learning work. Designed interaction improves distance education (Bernard et al., 2009); blended learning outperforms purely face-to-face teaching (Means et al., 2013); social presence is linked to satisfaction and perceived learning (Richardson et al., 2017).
  • Active work beats lecture. Active learning raised performance by about 0.47 SD; lecture students were 1.5× more likely to fail (Freeman et al., 2014).
  • Mixed groups, shared context. Same-ability grouping shows only small gains (Lou et al., 1996), while weaker learners gain from mixed groups (Saleh et al., 2005). Our design: grouped by area of work for relevance, mixed by role and experience for perspective.

Instructors: amplified, not replaced

  • In a randomised trial, AI assistance for tutors improved student mastery, most for less experienced tutors (Wang et al., 2024, preprint).
  • Generative AI cut lesson-planning time with no loss of quality in a teacher trial (Education Endowment Foundation, 2024).

What this means for security awareness

Field data cuts against the standard product in this market: annual click-through training shows little effect on behaviour (Ho et al., 2025; Lain et al., 2022; Prümmer et al., 2024). We don’t sell Sparkurity as a phishing-click reducer. We use role-based cases to build decision-makers’ skills — risk, incident response and audit.

References

Adaptive learning and tutoring

Mastery learning

Prior knowledge and expertise reversal

Retrieval practice, spacing and feedback

Relevance and case-based learning

Collaboration and cohorts

Online and blended learning

Security-awareness training

AI assistance for educators

  • Wang et al. (2024), Tutor CoPilot, preprint — arxiv.org/abs/2410.03017
  • Education Endowment Foundation (2024), ChatGPT lesson-preparation trial, report — link