The platinum-group element (PGE) contents of komatiites have been used to estimate the PGE contents of their mantle sources and investigate temporal variations in the PGE systematics of the mantle throughout Earth history. However, the use of mantle melts to investigate source characteristics is critically dependent on understanding the fractionation of the elements in question between mantle sources and their derived magmas. We present new PGE and Re abundance data, together with Re-Os isotope data for the pristine 1.9 Ga Winnipegosis komatiites, Canada. Whole rock Re-Os isotope data define a 1865 ± 40 Ma isochron, consistent with previous age estimates, with initial γOs of +0.9 ± 1.4. The PGE and Re variations in these komatiites were controlled by fractionation of olivine, chromite, Os-Ir alloys, a Ru-Os rich phase (possibly laurite) from the komatiite melt. The calculated parental melt composition at 23.6 ± 1.6 wt% MgO has 1.1 ± 0.1 ppb Os, 0.91 ± 0.08 ppb Ir, 4.0 ± 0.5 ppb Ru, 7.0 ± 0.8 ppb Pt, 7.2 ± 0.7 ppb Pd, and 0.53 ± 0.06 ppb Re. The calculated PGE contents are low relative to many komatiites; using existing methods for calculating the PGE contents of mantle sources suggests the Winnipegosis source had PGE contents from 25 ± 19 wt% to 57 ± 7 wt% of primitive mantle values. This surprising finding substantiates a growing ‘PGE paradox’, where almost all komatiites appear to have formed from PGE-depleted mantle sources. We suggest the simplest resolution to this paradox is that current methods systematically underestimate mantle source PGE contents. We test the assumptions implicit in these methods by comparing detailed models of PGE behaviour during mantle melting to a large database of parental melt compositions of high degree melts. Ruthenium behaviour in high degree melts can be adequately modelled using our current understanding of Ru partitioning. However, the Ru contents of high degree mantle melts are strongly dependent on the degree of melting, as Ru-rich melts formed after sulphide exhaustion are increasingly mixed into early-formed Ru-poor melts. Platinum and Pd behaviour during mantle melting cannot be successfully modelled under the widespread assumption that they are strongly incompatible following sulphide exhaustion, suggesting that these elements cannot be used to estimate PGE contents of the mantle sources of high degree melts. Parental melt Pt and Pd in many high degree lavas are correlated with Al2O3/TiO2, suggesting a pressure control on their partitioning. We attribute the relatively low PGE contents of the Winnipegosis komatiites to their moderate degrees and depth of melting compared to other komatiites, rather than invoking a PGE-depleted mantle source. More broadly, the complex controls (T, P, fO2, S content, prior depletion) on the PGE contents of komatiites mean that most komatiites can be reconciled with being derived from sources with primitive mantle-like PGE contents, and few komatiites can be confidently asserted to come from PGE-depleted mantle sources. Pending a better understanding of PGE behaviour during mantle melting, we recommend caution when using komatiite PGE systematics to infer the highly siderophile element evolution and accretionary history of Earth’s mantle.