Choose a Method

Default starting points

Situation	Recommended first method	Why
One interpretable seasonal signal	STD or STDR	simple configuration and stable outputs
One signal where component grouping matters	SSA	flexible component grouping with native support
Multiple aligned channels with shared structure	MSSA	joint decomposition across channels
You already know the main period	STL	reliable upstream baseline

Use the recommender when

• you want a machine-readable shortlist,
• you need to trade off speed versus accuracy,
• you want to exclude optional backends,
• you want to require native-backed methods.

detime recommend --length 192 --channels 3 --prefer accuracy

Move to wrappers when

• you specifically need adaptive decompositions such as EMD, CEEMDAN, or VMD,
• you need wavelet-based multi-scale structure,
• you need optional multivariate backends such as MVMD or MEMD.

Avoid starting with

• experimental wrappers before you have a stable baseline,
• multivariate methods on data that is really just one series,
• deprecated tsdecomp imports in new code,
• benchmark-derived methods in this repository, because they live in the companion benchmark repository.