Mechanistic language in molecular pharmacognosy often compresses several distinct evidentiary questions into a single claim: whether the phytochemical entity is correctly identified, whether a molecular target is merely predicted or physically engaged, whether perturbation of that target changes a relevant pathway, and whether the pathway contributes to the observed phenotype. This perspective examines why those questions should remain analytically separate and proposes a graded evidence ladder for natural-product mechanism claims. The analysis integrates phytochemical authentication, docking and target nomination, chemical-probe quality, chemoproteomics, target deconvolution, cellular target-engagement methods, network-level interpretation, and phenotype-oriented validation. The central proposal is that stronger mechanistic language should require closure of specific inferential gaps rather than accumulation of experiments that all interrogate the same level. Computational prediction can prioritize hypotheses but cannot substitute for target engagement; engagement can strengthen target attribution but does not establish functional necessity; pathway changes can demonstrate biological perturbation without proving target-specific causation; and phenotypic activity can remain compatible with polypharmacology, metabolites, mixture effects, or context-dependent biology. The proposed ladder is therefore intended as an interpretive and reporting framework, not a validated score or universal ranking system. Its purpose is to make unresolved links visible, preserve alternative explanations, and encourage mechanism statements whose strength is proportional to the evidence actually obtained. Important boundaries include uncertain phytochemical identity, assay-specific artifacts, indirect engagement signals, heterogeneous biological states, multiple active constituents, and incomplete exposure relevance.
