Scientific case study · interacty
Why iteration, not generation, is the real product in interactive learning content.
A peer‑style case study of interacty, EdTechLab's interactive‑content studio. It makes one argument with evidence: one‑shot AI generation is now commoditised, so the durable advantage is making the ten to thirty‑five rounds of refinement that actually produce a great artifact fast, trustworthy and cheap — and proving the preview is byte‑identical to the shipped SCORM package.
0.47 SD
active‑learning lift in exam performance over lecturing
Freeman et al., PNAS 2014 (225 studies)
33.8% → 21.8%
course failure rate, lecturing vs active learning (OR 1.95)
Freeman et al., PNAS 2014
≈ 92%
of real LMS course launches still use SCORM, not xAPI/cmi5
Rustici SCORM Cloud, 2025
δ = 0
preview‑to‑deliverable drift, by construction (§ The fidelity invariant)
interacty architecture
≈ 7×
cheaper to build a 35‑iteration deliverable ($3 vs $21 in AI)
structured‑diff cost model
10–35
iterations behind each real, deployed SCORM artifact
six Blackboard packages
The problem: a modern VLE is used as a filing cabinet
For most modules, a Virtual Learning Environment — Blackboard/Anthology Ultra, Canvas, Moodle, Brightspace — hosts a PDF where a clickable, trackable experience should be. That is not a cosmetic complaint; it is a pedagogical one with a measured cost. interacty exists to remove it. The demand is bottom‑up: this case study's artifacts began with a real institutional need at Edge Hill University — a careers and learning‑design team building a richer Blackboard template after seeing interactive labs with embedded quizzes, a forced‑sequence “Module Essentials” gate, a career‑pathway initiative, and a media‑rich student newsletter. The ask was never “more documents.” It was interactivity an ordinary academic could author and an institution could trust.
Why is interactivity worth engineering for?
Because four independent literatures agree that how a learner engages, not just what they receive, drives outcomes — and they give a quantitative target an authoring tool can be built against. The ICAP framework (Chi & Wylie, 2014) predicts a strict ordering by cognitive engagement:
Passive
Reading a PDF
Lowest learning gain
Active
Clicking, manipulating
Hotspots, sorting, timelines
Constructive
Generating, explaining
Self‑explanation: g = 0.55
Interactive
Co‑constructing, feedback
Highest predicted gain
The headline magnitude is large: across 225 STEM studies, active learning raised exam performance by 0.47 standard deviations and cut failure rates from 33.8% to 21.8% (Freeman et al., 2014); the gains disproportionately help under‑represented students (Theobald et al., 2020). Multimedia‑design effects give a precise grammar — contiguity g = 0.74, signalling g = 0.38, and, crucially, seductive details g = −0.41 (irrelevant “interesting” material actively harms learning). Retrieval practice lifts one‑week recall from 40% to 61% (Roediger & Karpicke, 2006), and gamification shows positive cognitive (g = 0.49) and motivational (g = 0.36) effects (Sailer & Homner, 2020).
Reported honestly: the macro‑direction (deeper engagement beats passive reception) is robust; the fine Interactive‑vs‑Constructive distinction is contested in the recent literature. The safe design rule is conservative — move content up several rungs (PDF → manipulable → generative), and capture the lift.
The thesis: “anything → interactive,” where the preview is the deliverable
Two words are load‑bearing. “Anything” — output beyond templated quizzes to bespoke, game‑like, even 3D interactivity. “Exact” — the previewer is the deliverable: what the author sees and interacts with is byte‑for‑byte what ships in the SCORM package. The mechanism is one architectural rule: the model edits a structured course graph; it is never the renderer. A single deterministic evaluator compiles the graph to a self‑contained HTML/CSS/JS artifact; rendering, preview and export are all functions of that one artifact. The model proposes typed, schema‑valid edit operations against addressable nodes — never final deliverable HTML on the default path.
Formal models
Three results that compose into one product decision.
Model 1 · Fidelity invariant
“What you see is what you ship” is a theorem, not a test.
δ = ε(M, L | C) → 0
Because preview and export run the same artifact, drift collapses to the conformance gap between the mock runtime and the real LMS. A traditional tool adds a second, divergent render path: δtrad = δpath + ε ≥ δinteracty. interacty removes δpath by construction.
Model 2 · Iteration ceiling
Cheaper iterations buy higher attainable quality.
Q(n) = Q★ − (Q★ − Q0)·ρn
Under a fixed author budget B, feasible iterations are ⌊B/c⌋, so quality rises as per‑iteration cost c falls. With free manual edits and ~7× cheaper AI edits, the same author reaches the ceiling instead of stalling: in a worked case, ~80% → ~100% of Q★.
Model 3 · Cost of an edit
A structured diff is ~20× cheaper than a regeneration.
cedit = pcin·s + pout·Δ ≪ cregen
Cached prefix (10% of input) plus a tiny patch Δ instead of re‑emitting the whole artifact L gives ≈20× per iteration, ≈7× per deliverable: a 35‑iteration flagship costs $3 vs $21, sustaining a ≈90% gross margin while manual edits stay free.
The evidence: six SCORM packages, deployed in Blackboard
The models are not hypothetical. Six self‑contained SCORM packages — authored by the researcher and running in Blackboard at Edge Hill University — instantiate the quality bar interacty is built to clear. All are self‑contained (the only external reference is a web font), mixing SCORM 1.2 and SCORM 2004 4th Edition as appropriate.
| Artifact | SCORM | Scale | Iterations |
|---|---|---|---|
| SPA Graduate Skills Attribute | 1.2 | 1,305 lines · 18 listeners · 6 quizzes | ≈10 |
| Declaration of AI Support | 1.2 | 1,378 lines · 58 cmi‑state markers | ≈10 |
| SCORM Lab — CIS1703 Wk 8 | 2004 4th | 308‑line runtime wrapper · localStorage preview fallback | ≈10 |
| Week 8 Tkinter lecture | 2004 4th | ≈1,770 lines interactive HTML | ≈10 |
| Week 10 package | 1.2 | ≈1,665 lines · 8 quiz/score structures | ≈10 |
| TechHub Student Experience | 1.2 | ≈105 MB · 137 images · 12 videos | 35+ (ongoing) |
The datum the whole architecture is built around: the quality bar is “10–35 iterations,” not “one prompt.”
The SCORM Lab's localStorage fallback —
the same artifact behaving correctly with and without an LMS — is the empirical seed of the mock runtime
that makes preview‑equals‑deliverable possible.
Questions, answered
Frequently asked
What is interacty, in one sentence?
interacty is EdTechLab's web studio for turning anything into a rich interactive experience — built from a block library or with optional AI — previewed exactly as it ships and exported as a self‑contained SCORM 1.2 / 2004 package (or standalone HTML) for any LMS.
How does “preview equals deliverable” work?
A single deterministic evaluator compiles the course graph to one self‑contained artifact. The preview runs that artifact against a faithful mock SCORM runtime; export zips the same artifact plus a generated manifest. Because there is no second rendering path, preview‑to‑deliverable drift is zero by construction (δ = 0) rather than something you test for.
Why does interactivity matter for learning?
Across 225 STEM studies, active learning raised exam performance by 0.47 SD and cut failure rates from 33.8% to 21.8% (Freeman et al., 2014, PNAS). The ICAP framework predicts that moving content up the engagement ladder — from passively reading to actively manipulating to generating — raises learning gain.
Why is iteration, not one‑shot generation, the focus?
By 2026, “upload a document → get a SCORM course” is commoditised. Real artifacts take 10–35 rounds of refinement. interacty makes those rounds cheap: manual direct‑manipulation edits are free to serve, and AI edits are structured diffs that cost ≈7× less per deliverable than full regeneration ($3 vs $21 for a 35‑iteration flagship).
Does interacty export SCORM for Blackboard?
Yes — self‑contained SCORM 1.2 and SCORM 2004, plus standalone HTML. SCORM still accounts for ≈92% of real LMS launches, and the six case‑study packages run in Blackboard today. Learner records stay under the institution's control via the LMS SCORM API; core data is EU‑hosted and GDPR‑ready by design.
What is proven, modelled, and imported
In the interest of trust: the fidelity invariant is an architectural property given a spec‑conformant mock runtime; the cost and iteration figures are modelled from verified per‑token prices and realistic action profiles, not yet measured over a production cohort; and the learning‑outcome effect sizes are imported from the cited primary literature — this case study argues interacty targets the right lever, it does not claim a measured efficacy trial of interacty‑authored content. A controlled study comparing the same module as a static PDF versus an interacty artifact is the natural next step.
See the product behind the case study.
interacty is part of EdTechLab's interactive‑content suite. Explore how it fits alongside intle and EngagedLab, or talk to us about a pilot.