A ground-truth-labelled dataset of 240 synthetic coupled ECG-photoplethysmogram (PLETH) signal pairs (60 s each, 500 Hz), varying coupling strength (alpha = 0 to 1), additive noise (SNR 5/10/20 dB and clean), and missing segments (0/10/30/50 percent), with 3 replicates per condition. Each case is a gzip-compressed float32 NumPy array (2 x 30000) with a manifest.tsv index carrying the ground-truth labels (alpha, snr_db, frac_missing, seed). Generated deterministically by the wavecouple package (DOI 10.5281/zenodo.23797941). Intended for benchmarking cross-signal coupling estimators (coherence, mutual information, Granger causality) under known conditions.