Heteroatom-doped metal-free carbon catalysts (HMCs) achieve high selectivity for the two-electron oxygen reduction reaction (2e⁻ ORR) to hydrogen peroxide (H2O2), yet mechanism assignments remain disputed. We identify three recurrent sources of disagreement: assuming dopant-centered active sites while the carbon–oxygen framework co-evolves; small, single-study sample sets amplified by within-element ratios; and inconsistent electrochemical and XPS reporting. To resolve this, we establish a cross-study evidence standard combining standardized electrochemical metrics, survey-anchored absolute atomic quantification, and DOI-grouped validation. Across alkaline studies, performance clusters in an narrow region of high sp² connectivity and moderate oxygenation (sp² fraction ≈82–88% and O ≈8–15 at%), associated with ≥90% H2O2 selectivity and favorable onsets. In an exclusive N-doped subset, nitrogen descriptors do not generalize under the same criteria. Machine learning-guided DFT on controlled sp²/sp³ frameworks with and without substitutional N corroborates the framework-first mechanism: the sp²/sp³ balance governs *OOH thermodynamics, whereas N produces only minor changes. Demonstrated here for alkaline 2e⁻ ORR on HMCs, the evidence standard is transferable to other mechanistically disputed catalytic systems.