Observational cohorts that report egg-associated cardiometabolic harm in diabetics cannot separate eggs from the confounders that co-travel with high egg intake (worse overall diet, socioeconomic and health-behaviour gradients, reverse causation). Randomisation breaks that confounding: with allocation to high- vs low-egg arms independent of baseline risk, a between-arm difference is attributable to the egg exposure itself. The trial was run in the very population the cohorts flag (prediabetes/T2D), so the test is population-matched rather than an extrapolation, and the outcome was pre-specified (change in HDL-C) with the result a tight CI straddling zero (+0.02 mmol/L, -0.03 to 0.08) plus consistent nulls across the other lipids and glycemic control. A pre-registered primary endpoint returning null is hard to reconcile with a real, large adverse effect being hidden by selective reporting. Two caveats bound the strength: (1) both arms were counselled to raise MUFA/PUFA and matched on macronutrients, so a genuine adverse egg effect could be partly offset by the shared background-diet shift, biasing toward the null; (2) 3 months and surrogate lipid/glycemic endpoints do not probe hard cardiovascular outcomes or long-term exposure. Net, the randomised null meaningfully lowers the probability that habitual high egg intake harms cardiometabolic risk factors in diabetics, without settling the hard-endpoint question.
Validity (step 6): approved / checked. Reconstructed step: randomisation makes arm allocation independent of baseline confounders, so a between-arm lipid/glycemic difference (or its absence) is attributable to the egg exposure, not to the diet/behaviour gradients that plague the cohorts; a pre-registered primary endpoint returning a tight CI around zero is therefore genuine unconfounded counter-evidence to the cohort harm signal for those surrogate endpoints. I traced the two undercutting defeaters the author already surfaces and confirmed neither breaks the step but both are correctly fenced in: (1) both arms were counselled toward MUFA/PUFA and macronutrient-matched, so a real egg effect could be partly offset — this biases the estimate toward the null, i.e. it caps how strongly the null reads, exactly as the statement says; (2) 3-month surrogate endpoints do not reach hard CV outcomes, so the conclusion is scoped to “cardiometabolic risk factors,” not CVD events. Because the conclusion is already the appropriately hedged “strong counter-evidence against the risk-factor harm signal, capped by the diet match / surrogate endpoints” — not “eggs are safe” — the inferential step holds as stated. (Per rule 1, that the argument is weak on hard endpoints does not downgrade validity; the step it actually makes is real.)