Choosing a DFT Protocol That Ranks Deep-Blue TADF Emitters and Pairs Them with Hosts

Computational screening of deep-blue thermally activated delayed fluorescence (TADF) materials decides which emitters are made and which host each is paired with, yet the descriptors behind these decisions are rarely tested against measurement. We test them on 50 molecules (39 published emitters, eight references and three phosphine-oxide hosts) against measured data for 27 emitters. The range-separated ωB97X-D3 functional overestimates every singlettriplet gap (∆EST) and fails to rank the emitters because its error differs between molecular families; a family offset restores the ranking, and the global hybrids PBE0 and B3LYP rank the gaps directly (Spearman ρ = 0.59 and 0.58). Rate constants built on computed gaps are unreliable, but become accurate once the measured gap is used. For host–emitter pairing, the triplet energies of host and emitter must be computed on one footing: a calibrated host combined with as-computed emitters misclassifies almost every confined pair. A tiered, costed protocol turns these findings into selection rules for screening campaigns.