We closely examine the trace-element concentrations of fifteen microinclusion-bearing diamonds from Canada, Guinea and South Africa that trapped low-Mg carbonatitic, silicic and saline high-density fluids (HDFs). The microinclusions trapped only HDFs, with no mineral microinclusions or mixtures of HDF+mineral; thus, their LA-ICP-MS analyses solely represent the compositions of the trapped fluids.HDFs of different major-element compositions (silicic, carbonatitic and saline) are characterized by fractionated REE patterns and variable, mostly negative, anomalies in Sr, Ti, Zr, Hf and Y relative to REEs of similar compatibility, regardless of their host diamond's provenance. In the highly incompatible elements (Cs-Pr) two patterns are notable. "Ribbed" patterns are characterized by high levels of Ba, Th, U and LREEs and lower alkalis, Nb and Ta. "Planed" patterns are smoother and typically devoid of significant fractionation between elements of similar compatibility. Co-variation diagrams of (La, Ce)/(Nb, Rb) vs. (U, Th)/(Nb, Rb) and Rb/Nb vs. La/Nb ratios can be used to best distinguish between the two patterns.Similarities of canonical ratios, such as Nb/(Th, U, La) and K/U, between MORB and OIB samples and HDFs with "Planed" patterns suggest an asthenospheric source for these HDFs. For silicic compositions, this idea is strengthened by calculating the sources in equilibrium with such HDFs which range in composition between the DMM and more fertile parts of the convecting mantle. Isotopic analyses may pinpoint this connection.The trace-element patterns of MARID and PIC xenoliths show a mirror-image to the pattern of silicic, low-Mg carbonatitic and saline HDFs with "Ribbed" patterns. Assuming that the HDFs are the product of a 0.1% batch melting, we obtain a source that is similar to the MARID and PIC patterns. However, the decoupling between major- and trace-elements in the HDFs argues against melting or fractional crystallization as the main processes leading to the formation of the "Ribbed" patterns.Percolation of an asthenospheric silicic HDF with "Planed" pattern through previously metasomatized lithosphere that carries accessory phlogopite and Fe-Ti oxides, closely reproduce the "Ribbed" pattern of silicic and low-Mg carbonatitic HDFs at fluid/rock ratios ≈ 0.1%. The initial trace-element pattern of the lithosphere influences the more compatible elements of the HDF (Sr-Lu). However, in the Cs-Pr range, the presence of phlogopite and Ti-Fe oxides controls the evolution of the "Ribbed" pattern. Percolation explains the observed decoupling between major- and trace-elements in HDFs and the resemblance of trace-element patterns in HDFs from different cratons. It may also explain the limited variation of δ13C in fibrous diamonds (-6 ± 2‰). The two patterns escape the circular "chicken and egg" reasoning that calls for an enriched source for the formation of highly fractionated melts: it suggests that diamond-forming fluids can come directly from the asthenosphere (with no need for a pre-metasomatized source) and that they can be further modified in the lithosphere.