Diminishing returns are the expected shape, not a failure
Adaptation curves flatten. The first year of consistent training produces changes that the fifth year cannot, and this is a property of the biology rather than a sign that something is being done wrong. A great deal of optimisation content sells the flattening itself as a problem with a purchasable solution.
Interpreting your own plateau starts with knowing what the achievable remaining range is. In most trained populations the residual effect sizes available from any single intervention are small — which is precisely why they are hard to detect in your own data, and easy to claim in marketing.
Reading a study: the four things that decide what it is worth
Species first. A result in mice is a hypothesis about humans, not a finding in humans. This single distinction disqualifies a large share of what gets cited in optimisation circles, and it is why every study summary on this site is labelled.
Then sample size and duration. A twelve-person, six-week crossover can detect a large effect and is nearly blind to a small one. Absence of a significant result in a small study is not evidence of absence.
Then the endpoint. What was actually measured is frequently not what the abstract's framing implies — a marker moved, which is not the same as an outcome changed. Surrogate endpoints are where most overclaiming lives.
Then the comparator. Against placebo, against active control, or against nothing at all. Uncontrolled trials in this space are common and tell you very little.
Effect size beats statistical significance
A p-value tells you how surprised to be if there were no effect. It does not tell you how large the effect is, and with a large enough sample, trivially small differences become statistically significant.
The number worth extracting is the effect size and its confidence interval. A confidence interval that spans from 'negligible' to 'meaningful' is an honest result reported honestly, and it is far more common in this literature than the summaries suggest.
Where the evidence genuinely runs out
For most compounds discussed in optimisation contexts, there is no long-term human safety data at all, because the trials that would produce it have not been run. This is a statement about the state of the literature, not a claim in either direction about risk.
There is also very little research on healthy, already-optimised populations. Most trials recruit people with a deficit to correct, and effects observed while correcting a deficit do not transfer to someone without one. This is one of the most consistently ignored limitations in the field.
What this research does not establish
- Almost no published research studies healthy, already-optimised people. Extrapolating from deficit-correction trials to this population is not supported.
- Long-term human safety data does not exist for most compounds in this literature.
- Effects that are real but small are difficult to distinguish from noise in individual self-experiments, however carefully they are run.
Common questions
- I did everything right and I'm still stuck — what does the research say?
- That flattening returns are the expected shape of an adaptation curve rather than evidence of a correctable error. The research relevant to already-optimised people is thin, because most trials recruit participants with a deficit to correct, and results from correcting a deficit do not transfer to someone without one.
- I've maxed out diet and training — what does the literature actually support next?
- Less than the surrounding content implies. For trained populations the remaining effect sizes from any single intervention are generally small, and small effects are exactly the ones that individual self-experiments cannot reliably detect. Reading the primary literature on effect size and confidence intervals is more useful than another protocol.
- What does 'beyond the basics' optimization actually mean?
- In practice it means the interventions with the largest, best-evidenced effects have already been applied, so what remains are candidates with smaller effects and thinner evidence behind them. The research base narrows sharply at that point: most trials recruit people with a deficit to correct, and very little work studies healthy, already-optimised populations. Reading primary literature on effect sizes becomes more useful than looking for another protocol.
- How do I tell a good study from a bad one?
- Check four things in order: species (human or animal), sample size and duration, what endpoint was actually measured rather than what the framing implies, and what it was compared against. A large-sounding claim from a small, short, uncontrolled animal study is common in this field.
- Is there long-term safety data for the compounds discussed in this space?
- For most, no — the long-duration human trials that would produce it have not been conducted. That is a description of the evidence base, not a claim that anything is safe or unsafe. Absence of data is not the same as evidence of safety.