Only 2 of 14 carb loading studies were properly blinded, and both found nothing

What a double blind placebo controlled trial is

In a double blind placebo controlled trial neither the participant nor the person measuring them knows which condition they are in, so what someone believes about the treatment cannot leak into the result.

It is unusually hard to arrange for a diet. Eating nine or ten grams of carbohydrate per kilogram a day for three days is difficult to hide from the person doing the eating, which is why so little of this literature manages it.

Drawn from background method, not from this paper.

Why they ran it

Carbohydrate loading has long been a cornerstone of endurance nutrition, the authors write, and yet its true efficacy remains uncertain. Belief and expectation can move endurance performance on their own.

So they asked a question about the shape of the evidence rather than about carbohydrate: how much of the loading literature was built in a way that could tell the diet apart from the belief in the diet.

Drawn from the paper's abstract and stated aims.

Fourteen carbohydrate loading studies with a performance outcome were identified and evaluated. Only two were double blind and placebo controlled trials. Only four provided any carbohydrate during the exercise test itself.

Both of the double blind placebo controlled studies also included appropriate carbohydrate intake during exercise, and both reported no evidence of a performance benefit from loading.

Of the remaining two studies that provided carbohydrate during exercise, loading did show a positive performance effect, but the authors note the effect size was moderate and comparable to what has been reported for placebo nutritional interventions, a Cohen's d of about 0.35.

The trials are also heterogeneous in what they measured, some using time to exhaustion and some using a time trial, which the authors say points to a need for standardised conditions.

The numbers

Studies with a double blind placebo controlled design	2 of 14
Studies providing carbohydrate during exercise	4 of 14
Performance benefit in the two blinded studies	none found
Effect size in the two fuelled but unblinded studies	Cohen's d about 0.35, comparable to placebo nutritional interventions
Duration window where loading is usually claimed to help	about 90 to 240 minutes

Why this might happen

Proposed by the authors This is the explanation the authors offer in their discussion. This study did not test it.

The mechanism the authors are pointing at is not metabolic. An athlete who knows they have loaded expects to perform better, and expectation on its own moves endurance performance by an amount similar in size to the benefit that unblinded loading trials report.

Their second point is about the race rather than the athlete. When carbohydrate is taken during the event, starting with a fuller muscle matters less, which is why they argue the whole question has to be asked again with in race fuelling optimised.

Drawn from the paper's argument.

What this does not show

It does not show that carbohydrate loading does not work. Two blinded trials finding nothing is not the same as nothing being there. The authors' claim is that the literature cannot currently answer the question once expectation is accounted for, and they call for the trials that would.
No effect sizes were pooled. This is a commentary with a systematic evaluation of study designs, not a meta analysis. It counts how the trials were built and what they concluded; it does not combine their results into a single number.
In race fuelling replacing loading is an argument, not a tested comparison. The reframing toward carbohydrate during exercise is the authors' reading of why the blinded trials came out flat. No head to head trial of loading against in race fuelling is reported here.
Blinding is not the only thing separating these trials. Time to exhaustion and a time trial are different tests that reward different things. Design differences beyond blinding remain in the mix and are not resolved by counting blinded studies.

Where this leaves us

The performance case for carbohydrate loading rests on a literature that mostly could not separate the diet from the belief in the diet. Twelve of fourteen trials were not blinded, and the two that were, both of which also fuelled during exercise, found nothing. This does not disturb the glycogen literature, which is solid and consistent. It puts a question mark on the step from glycogen to speed.

The published carbohydrate loading performance literature in endurance exercise as the authors identified it, which is fourteen studies.

Double blind placebo controlled trials with carbohydrate provided during exercise, which is what the authors explicitly call for, run on standardised performance tests so the trials can be compared with each other.

Caveats worth holding

A commentary with a systematic design evaluation, not a meta analysis. It reports that the evidence base is thin, not that loading has been shown to fail.
Absence of evidence in two small blinded trials is not evidence of absence.
The count of fourteen is the set of studies the authors identified with a performance outcome, and a different inclusion rule would produce a different denominator.