Supplementary Materials

This supplementary package accompanies the paper and provides the main resources underlying the study, with particular focus on the final annotation codebook and the derived datasets.

Contents

1. codebook_final.pdf

Final annotation codebook used in the study. It includes:

• the full set of annotation labels
• their operational definitions
• inclusion and exclusion criteria
• temporal thresholds where applicable
• the reduced set of final labels used for baseline modeling

2. dataset_overview.pdf

Overview of the released datasets and their construction pipeline. It summarizes:

• the distinction between CAM1, CAM2, and multi-view 3D datasets
• the temporal windowing strategy
• the final target labels
• the selection criteria used to build the ML-ready subsets

3. data_dictionary.csv

Column-level description of the released datasets, including:

• variable name
• type
• description
• dataset applicability

4. thematic_analysis.pdf

Document reporting the thematic analysis table derived from the educator interviews and used to support the co-design process.

5. datasets/

Derived tabular datasets used for baseline modeling.

datasets/cam1/

Person-level 2D feature dataset derived from CAM1.

datasets/cam2/

Person-level 2D feature dataset derived from CAM2.

datasets/multiview_3d/

Final person-level multi-view 3D datasets derived from real triangulation and controlled fusion.

6. results/

Summary files reporting the main baseline results for the released datasets.

Notes on data format

• Each row in the released datasets corresponds to a person-level temporal window.
• The datasets do not contain raw videos or directly identifying information.
• Annotated identities are pseudonymous and used only as internal linking variables across derived representations.
• The released files are derived feature representations designed for analysis and baseline modeling.

Ethical note

The supplementary materials do not include raw audiovisual recordings. The released datasets are limited to derived, non-biometric representations consistent with the privacy-preserving design of the project.

Relationship between files

• The annotation codebook defines the labels.
• The dataset overview explains how labels and features were converted into trainable datasets.
• The data dictionary explains the meaning of each column in the released CSV files.