Description: Highlights • Subduction plate interface constitution, thickness, geometry and viscosity • Rocks recovered are reliable probes of subduction processes and thermal regimes • Long-term mechanical coupling controls rock recovery from subduction zones • Subduction cooling induces a long-term trend in rock detachment from the slab • Rock recovery is a proxy for long- and short-term subduction dynamics and coupling Abstract Short- and long-term processes at or close to the subduction plate interface (e.g.,mineral transformations, fluid release, seismicity and more generally deformation) might be more closely related than previously thought. Increasing evidence from the fossil rock record suggests that some episodes of their long geological evolution match or are close to timescales of the seismic cycle. This contribution uses rocks recovered (episodically) from subduction zones, together with insights from thermomechanical modelling, to provide a new dynamic vision of the nature, structure and properties of the plate interface and to bridge the gap between the mechanical behavior of active subduction zones (e.g.,coupling inferred from geophysical monitoring) and fossil ones (e.g.,coupling required to detach and recover subducted slab fragments). Based on critical observations and an exhaustive compilation of worldwide subducted oceanic units (for which the presence near the plate interface, rock types, pressure, temperature, T/P gradients, thickness and timing of detachment can be assessed), the present study demonstrates how long-term mechanical coupling exerts a key control on detachment from the slab and potential rock recovery. Critical assessment of rock T/P characteristics indicates that these fragments can indeed be used as natural probes and provide reliable information on subduction interface dynamics down to ~2.8 GPa. Rock clusters are identified at depths of 30, 55–60 and 80 km, with some differences between rock types. Data also reveal a first-order evolution with subduction cooling (in the first ~5 Myr), which is interpreted as reflecting a systematic trend from strong to weak mechanical coupling, after which subduction is lubricated and mostly inhibits rock recovery. This contribution places bounds on the plate interface constitution, regular thickness (〈300 m; i.e. where/when there is no detachment), changing geometry and effective viscosity. The concept of ‘coupled thickness' is used here to capture subduction interface dynamics, notably during episodes of strong mechanical coupling, and to link long- and short-term deformation. Mechanical coupling depends on mantle wedge rheology, viscosity contrasts and initial structures (e.g.,heterogeneous lithosphere, existence of décollement horizons, extent of hydration, asperities) but also on boundary conditions (convergence rates, kinematics), and therefore differs for warm and cold subduction settings. Although most present-day subduction zone segments (both along strike and downdip) are likely below the detachment threshold, we propose that the most favorable location for detachment corresponds to the spatial transition between coupled and decoupled areas. Effective strain localization involves dissolution-precipitation and dislocation creep but also possibly brittle fractures and earthquakes, even at intermediate depths.