Testing specific predictions of competing hypotheses to explain conflicting gene trees. A. Patterned after an explicit test of the chromalveolate model of plastid evolution , the two hypothetical scenarios show mutually exclusive predictions about "algal" genes in a heterotrophic protist's genome based on two alternative evolutionary hypotheses. The upper scenario of a more ancient endosymbiosis predicts that genes from the secondary endosymbiont, those unrelated to plastid function, should be shared between the aplastidic heterotroph and its algal sister taxon. The lower scenario of a later, taxon-specific plastid origin, predicts that "algal" genes from the heterotroph are products of common HGT or phylogenetic artefacts and, therefore, should not be shared with the photosynthetic neighbor relative to a negative control (C-). The control is a taxon generally agreed to be unrelated, phylogenetically or through endosymbiosis, to either the host or endosymbiont lineages. If the shared phylogenetic signal from the putative endosymbiont is not significantly greater than from the control group, then there is no objective basis for advancing EGT as an explanation for apparent "algal" genes in the heterotroph's genome. B. EGT versus HGT in a heterotrophic taxon. In this case, a rigorous test could be whether there are significantly more "algal" genes in the organism of interest than in phagotrophic control taxa with no presumed history of EGT. If there is not a significantly greater signal of HGT from the presumed endosymbiont in the target genome than in the control taxa, then algal genes are consistent with HGT or phylogenetic artefacts and EGT is not supported. C. The same approach could be used to test whether repetitive HGT is a superior hypothesis to phylogenetic artefacts by examining control taxa that should have had little to no opportunity to take up DNA from the organism in question, based on their presumed ecological and evolutionary histories. If it is biologically unreasonable to expect common products of HGT in the control genome (C-), and there is comparable signal present as in the target genome (T), then HGT does not rise above the null hypothesis of signal from statistical biases and/or noise across genome-level data.