Headlines about a single food being harmful and then beneficial recur so often that the field appears unstable. The pattern comes from how nutrition can practically be studied.
Randomised trials are largely impossible
The strongest evidence comes from assigning people at random to different treatments, which for diet would mean controlling what thousands of people eat for decades.
Participants cannot be blinded to their food, compliance drifts over long periods, and the outcomes of interest often take many years to appear.
Short feeding trials are feasible and measure surrogate markers rather than the illnesses researchers actually care about, which limits what they can establish.
Observational data cannot separate the food from the eater
Most nutrition findings come from following large groups and comparing those who eat more or less of something, which records association rather than cause.
People who eat a particular food differ systematically in income, exercise, smoking, sleep and healthcare access, and every one of those affects the outcome.
Statistical adjustment reduces the problem without eliminating it, since unmeasured differences remain and researchers must choose which factors to adjust for.
Intake is measured by asking people
Dietary data usually relies on questionnaires asking participants to recall what they ate over a period, sometimes a whole year.
Recall is imperfect, portion sizes are estimated, and reporting skews towards what people believe they should be eating rather than what they did eat.
Errors of that kind blur real effects and can manufacture apparent ones, particularly when the effect being sought is small.
Small effects invite many analytical choices
Large datasets can be analysed in many defensible ways, and different choices about categories, adjustments and exclusions produce different results from identical data.
Findings that reach significance are more likely to be published and far more likely to be covered by news outlets, so the visible literature is not the whole literature.
Single studies therefore deserve much less weight than the accumulated pattern across many, which is how expert bodies actually form advice.
The broad conclusions have been stable
Underneath the churn, guidance on overall dietary patterns has changed remarkably little, while the claims about individual nutrients have swung repeatedly.
That difference is informative: patterns are measured more robustly than single components, and they are less sensitive to any one methodological choice.
Anyone with a specific medical condition or dietary concern needs individual advice from a qualified professional rather than inference from a headline.