Table 3.

Top two globally fixed configurations by pipeline and dataset.

Pipeline	Rank	Stieger2021		BNCI2014-004
		Configuration	Acc. (%)	Configuration	Acc. (%)
MAP	1	TS/LDA/CORAL	69.43	CSP/SVM/PT	74.52
	2	TS/LDA/none	69.43	CSP/LDA/SA	74.43
DWP	1	CSP/LDA/SA	67.70	CSP/LDA/SA	74.22
	2	CSP/SVM/SA	67.37	CSP/LDA/PT	73.83
MMPmta	1	TS/LDA/CORAL	68.63	CSP/LDA/SA	73.92
	2	TS/LDA/none	68.59	CSP/LDA/PT	73.59
MMPmoe	1	CSP/LDA/SA	67.22	CSP/LDA/SA	74.03
	2	CSP/LDA/PT	65.92	CSP/LDA/PT	73.72
BDP	1	TS/LDA/CORAL	68.63	CSP/LDA/SA	73.92
	2	TS/LDA/none	68.59	CSP/LDA/PT	73.59

Configurations are shown as feature/classifier/domain-adaptation triples. For each pipeline, each fixed configuration is first averaged across target folds within each subject and then averaged across subjects to rank the resulting global averages. Table 2 instead first computes the mean accuracy for each configuration across target folds within each subject, extracts the maximum accuracy within each subject, and finally averages these subject-specific maxima across subjects; consequently, the two aggregation sequences do not necessarily yield the same numerical value. Accuracy values are expressed as percentages. Throughout, BDP denotes the Bridge → Far proxy variant retained for the main analysis.