Data

Three CSV files, all linked by subject_id, all generated together by data/generate_datasets.ipynb:

File	Type	Columns	Used by
dataset_NPSY.csv	Synthetic	age, sex, criq, apacs, att_matr, moca	Case Study 1 & 2
dataset_ERP.csv	Synthetic	n400_amp, n400_lat, p300_amp, p300_lat	Case Study 1
dataset_PSD.csv	Real	mean_power, delta_power, theta_power, alpha_power, beta_power, gamma_power	Case Study 2 (summary only — the real analysis uses the full arrays below)

Each file has 45 rows (46 with the header): 40 clean participants used in the final analyses, plus 5 deliberately added with missing or implausible values, so each case study has something real to detect and exclude (see its Data Wrangling section). The audit trail (data/EEG/PSD/subject_id_mapping.csv) flags these five with a -dup suffix, since they are duplicated spectra rather than independent recordings.

What's synthetic and what's real

• Synthetic

  ---

  dataset_NPSY.csv and dataset_ERP.csv are entirely invented — so that the associations each case study demonstrates are known in advance, and so the tutorial can be run end to end without real patient data.

• Real

  ---

  Case Study 2's EEG spectra are real, anonymised recordings from the PATHS study (data/EEG/PSD/psd_matrix.npy, psd_log_matrix.npy, computed by data/EEG/PSD/compute_psd.ipynb from real epoched recordings), authorised for publication here. Case Study 2 itself reads the _full variants of these two arrays (psd_log_matrix_full.npy, psd_matrix_full.npy), which also include the 5 duplicated spectra mentioned above — the original 40-recording files are never modified.

Important disclosure

Because the real EEG and the synthetic neuropsychological table do not come from the same patients, the (synthetic) moca column was deliberately linked to a real, montage-based property of each recording. The resulting brain–behaviour association in Case Study 2 should not be read as a real clinical finding — see the notice at the top of generate_datasets.ipynb for the full disclosure.