Semax Before And After: What Was Measured, and In What

Someone posts a semax before and after: two runs of an online cognitive test three weeks apart, the second one better, with a note about feeling sharper at work. The comparison is honest in intent. It is also uninterpretable, and not because the person was careless.

The second measurement contains everything that changed between the two dates. The compound is one item on that list, and several of the others reliably push a repeated cognitive score upward on their own.

Semax before and after: what the second score contains besides the compound

Any paired measurement is a difference, and a difference attributes to nothing in particular. Between two testing sessions a person has practised the test, formed an expectation, slept differently, changed their caffeine intake, moved through a work cycle, and been selected into the comparison by the fact that they cared enough to run it twice.

An experiment separates those by holding a comparison condition alongside the exposed one, so that everything except the compound applies to both groups. Remove the comparison group and the arithmetic no longer isolates anything. This is not a criticism of amateur record keeping. A pharmaceutical company with unlimited resources gets the same uninterpretable result from an uncontrolled paired comparison, which is why nobody runs one on purpose.

Practice effects: repeated cognitive tests rise on their own

This is the specific reason cognitive outcomes are worse than most for paired records.

Cognitive tasks are learnable. On a second exposure a person knows the format, the timing and often the strategy, and scores improve for that reason alone. The effect is a well recognised nuisance in cognitive testing, which is why trials use parallel test forms, practice sessions before the measurement that counts, and a control group that also takes the test twice.

An individual comparing two attempts at the same task has none of those safeguards. The improvement in the second attempt is exactly what would be predicted with no intervention at all, and the size of the improvement is not a signal of anything, because nobody knows what the compound free version of that improvement would have looked like for that person.

Self administered tests amplify the problem. Effort is not fixed. Someone running a test to see whether something worked applies more attention on the second run than on the first, and attention is a large part of what these tests measure.

Self rated focus follows mood, and mood moves for its own reasons

Most posted comparisons are not test scores at all. They describe concentration, motivation and mental clarity, which are self ratings.

Self rated cognition correlates with current mood more strongly than it correlates with measured cognitive performance. A person in a better week reports thinking more clearly whether or not their measured performance changed. Anyone who has decided to try a compound has an expectation attached, and expectation moves self report reliably and in the direction expected.

That is the mechanism behind the striking consistency of positive reports online. It is not necessarily that nothing happened. It is that this instrument cannot distinguish something happening from expecting something to happen, and the more strongly a person anticipated an effect, the less their report can settle the question.

Regression toward the middle explains the best accounts

People do not begin an intervention at a random point. They begin during a bad patch: a period of poor focus, low mood, or unusual fatigue.

Extreme states tend to be followed by less extreme states regardless of what is done about them, because the extreme reading partly reflected temporary conditions that pass. Anything administered at the low point is credited with the return toward the person's usual level. This effect is strongest for exactly the outcomes people report here, since attention and mood fluctuate widely within a person over weeks.

The consequence is counterintuitive and worth stating: the more dramatic the reported improvement, the more likely the starting point was an outlier, and the less the comparison can tell anyone about the compound.

Four kinds of record, and what defeats each

The rodent row is included because it is the one kind of paired measurement on this compound that was made under controlled conditions. Studies in rats have compared exposed and unexposed animals on behavioural tasks and on biochemical markers in brain tissue, with the comparison built into the design. That is a proper before and after in the sense that matters, and it is a statement about rats.

Clinical records exist, and they cannot be audited from outside

There is a further category that most writing on this compound misses in one direction or the other.

This compound has been used medically in Russia under a national approval, which means clinical assessments were made, before and after treatment, in patients. That is human paired data collected by clinicians. It is also not available to a reader outside that system in the form needed to evaluate it, since much of the relevant literature is published in Russian in journals that are not comprehensively indexed in the databases most readers search, and the design details that determine how much a paired clinical result is worth cannot be checked.

Both common conclusions are wrong. Reporting that no human data exists is false. Presenting that data as equivalent to a published, registered, controlled trial is also false. What a careful reader can say is that human clinical assessment happened, that its quality cannot be inspected from outside, and that no Western trial exists to cite in its place. No trial registry identifier belongs in a discussion of this compound.

There is also a filter operating before any of this. Records that get posted are records where something happened. A person who tried the compound, noticed nothing, and stopped after a week rarely writes it up, and nobody screenshots an unchanged score. The visible collection of paired records is therefore not a sample of outcomes; it is a sample of interesting outcomes, and its apparent consistency is produced by the filter rather than by the compound.