HomeResearch Results

DESCRIPTIVE PILOT EVIDENCE

Evidence, with the context it deserves.

A small, short-duration pilot explored component workflows and learning outcomes. Its findings are promising observations that require broader validation.

50unique learners in total
25Ages 9–12 · the same learners evaluated C1 & C2
25Ages 6–8 · the same learners evaluated C3 & C4

TECHNICAL & WORKFLOW OUTCOMES

Different tasks.
Different measures.

No single “overall AI accuracy” combines
these component-specific results.

C1 / Geometryn = 25
25/25

agreement with expected validity

25 evaluated cases: four valid and one invalid case for each of five solids. All 20 valid cases progressed to 3D/AR; all five invalid cases stopped before reinforcement.

Mean task time: 12.65 s

Component details →
C2 / Geometryn = 25
80%

Virtual Slicing completion · 20/25

20 of 25 slicing sessions succeeded. Direction outcomes were 5/5 vertical, 11/14 horizontal and 4/6 angled. Fraction outcomes were 4/5 for 1/2, 3/4 for 1/3, 6/7 for 1/4 and 7/9 for 3/4.

Mean slicing duration: 6.40 s

Component details →
C3 / Literacyn = 25
76%

correct recognition · 19/25 trials

19 correct predictions, three incorrect predictions and three failed letter returns. Mean total session time was 8.35 seconds; mean drawing time was 5.05 seconds.

Mean total session time: 8.35 s

Component details →
C4 / Literacyn = 25
68%

full vocabulary / AR flow · 17/25

20/25 spelling completions (80%); 21/25 pronunciation-oriented acceptances (84%); 17/25 completed both tasks and unlocked AR (68%).

Mean total session time: 56.52 s

Component details →
Component metrics — manuscript Table II, page 5
ComponentMetricObserved outcomeMean time
C1agreement with expected validity25/25 · n=2512.65 s
C2Virtual Slicing completion80% · n=256.40 s
C3correct recognition76% · n=258.35 s
C4full vocabulary / AR flow68% · n=2556.52 s

C4 supplementary workflow outcomes: spelling 20/25 (80%); pronunciation-oriented acceptance 21/25 (84%); both tasks and AR unlocking 17/25 (68%). Confidence is not classification accuracy; pronunciation acceptance is not independently measured speech-recognition accuracy.

LEARNING OUTCOMES

Four marks higher.
In every component mean.

Pre/post averages on a ten-mark scale.
A 40 percentage-point descriptive change.

Pre-test Post-test
Aggregate pre/post means — manuscript Table III, page 5
ComponentAge groupPre-test / 10Post-test / 10Change
C19–1259+4 marks
C29–1259+4 marks
C36–859+4 marks
C46–848+4 marks

INTERPRETING THE PILOT

What remains
to be explored.

Only aggregate pre/post means were retained in the final analysis. The results do not establish statistical significance or a causal learning effect.

  • Small samples and a short study duration limit generalization.
  • C1 results describe agreement on 25 evaluated cases.
  • C2 evidence quantifies slicing rather than separate correction accuracy.
  • C3 sampled 17 uppercase targets; a complete alphabet benchmark is still needed.
  • C4 intended-object labels were not independently retained, so child-sketch classification accuracy cannot be calculated from confidence.
  • Longer retention studies, broader validation and a dedicated child-usability instrument remain future work.

The manuscript reports institutional ethics procedures, parental/guardian consent, learner assent and anonymization (page 3). No identifiable child testing photographs are republished here.

Read the research manuscript ↗

Supplied research manuscript; publication or acceptance has not been verified.