Timing Nutrition Changes: 3-Week Adjustments and the No-Overcorrection Rule

Timing Nutrition Changes: 3-Week Adjustments and the No-Overcorrection Rule

AI coaching works best when it changes nutrition on the same clock the body changes on: slowly enough to see signal, fast enough to stay ahead of drift.

The 2022 Carbohydrate Diet Study adjusted carbs by 100 g/day for three weeks and used the result to set the next move; that’s a clean example of the timing principle built into the broader mechanism of glycemic settling. When you change nutrition, the useful question is not just what to change, but when to change it again. For coaches using AI, the sharp thesis is this: if you adjust nutrition faster than the body can show a real response, you stop coaching physiology and start chasing noise.

The core mistake: reacting to every swing

In practice, most bad nutrition coaching comes from one of two errors. The first is waiting too long and letting momentum get away from you. The second is the more common AI-era problem: updating too often because the dashboard makes every day look meaningful.

That’s especially dangerous in physique coaching, where short-term changes are dominated by water, sodium, bowel content, stress, travel, sleep, and training inflammation. A scale jump or a flat morning does not automatically mean the plan failed. It means the system has more than one input.

Justin Harris’s coaching language in the KB is blunt on this point. He notes that during prep, when water intake dropped slightly on a drive back from KC, he ended up backed up for a couple days; in the same thread he connects constipation and distension to lower water intake and to the gut drying out during prep. His practical observation is not that every fluctuation needs a new plan. It’s that small changes can show up as digestive noise long before they show up as meaningful body-composition change. That is exactly why over-adjustment is so costly.

What the timing rule looks like

The evidence-aware version of nutrition timing is simple:

1. Make one change.
2. Hold long enough to see a real response.
3. Only then decide whether the change worked.

The 2022 Carbohydrate Diet Study gives the timing logic in miniature: a 100 g/day carb adjustment over three weeks. That is not a license to copy the exact protocol for every athlete. It is a reminder that meaningful nutrition feedback usually needs a real window, not a couple of noisy days.

For coaches, the practical takeaway is that the interval between adjustments should be long enough to let the main signal emerge. If the change is supposed to affect bodyweight, that signal may be obscured by digestion and hydration for several days. If it is supposed to affect fullness, performance, or recovery, you still need enough time to see whether the athlete settles into the new intake rather than simply lurching around it.

Why over-adjustment happens

AI makes it easy to think in terms of continuous optimization. But coaching is not a live recommender system. It’s a controlled experiment with an athlete attached.

Over-adjustment usually comes from reading the wrong layer of data:

• Day-to-day bodyweight instead of the rolling trend
• One flat training session instead of repeated performance pattern
• One digestively bad day instead of the week’s overall tolerance
• A single report of “I feel off” instead of a pattern across multiple check-ins

The more granular the data stream, the more tempting it is to intervene. But granularity does not equal truth. It can just mean more places to overfit.

That matters because nutrition changes often have delayed and layered effects. A carb increase may alter training output before it changes scale weight. A food-source change may improve GI comfort before it changes adherence. A reduction in intake may show up first as less fullness or worse pumps, while actual fat loss takes longer to declare itself.

So the coach’s job is not to eliminate uncertainty. It is to avoid mistaking uncertainty for failure.

The better way to time a change

A good adjustment rule in AI-assisted coaching should answer three questions before it acts:

1) What problem am I trying to solve?

Is the issue fat loss stalling, digestion worsening, performance dropping, or adherence slipping? Each one has a different time course.

If the problem is GI distress, you may need a faster response, but even then the fix should be targeted before it becomes systemic. Justin’s prep comments about water intake and constipation point to a very specific principle: when the body is “backed up,” the scale and waist can mislead you. A coach who keeps cutting calories in response to that may create a second problem while trying to solve the first.

2) What is the smallest change that could work?

The worst nutrition updates are the ones that try to solve too much at once. If you reduce carbs, fats, and meal frequency simultaneously, you won’t know which lever mattered.

The point of small changes is not caution for its own sake. It is attribution. A coach who changes one variable at a time can actually learn from the response. A coach who changes everything learns nothing.

3) How long until the signal is likely to appear?

That window depends on the variable. Digestion may respond quickly. Bodyweight trend takes longer. Full-body appearance can lag behind both. If you collapse all of those into a 24-hour feedback loop, you’ll over-steer.

The 3-week carb adjustment example matters because it illustrates a patience horizon. It tells us not to confuse “no immediate dramatic response” with “no response.”

What AI should do instead of guessing

AI is useful when it forces consistency. It can track check-in cadence, flag when the same issue repeats, and keep a coach from making emotional decisions in the moment. But it should not be allowed to spin every new data point into a plan revision.

A better AI workflow is:

• collect multiple days of data,
• compare against the prior trend,
• identify whether the issue is trend-based or noise-based,
• propose one narrow change,
• then wait for the next observation window.

That approach makes AI a pacing tool, not an impulse engine.

It also fits the reality of coaching experience. Justin’s comments in the KB are not about micromanaging every fluctuation; they are about recognizing predictable sources of temporary distortion, especially in prep. The same logic extends to nutrition timing more broadly. If you know certain inputs can create short-lived noise, you stop treating every short-lived change as an emergency.

The falsifiable rule

Here’s the thesis in plain terms: nutrition changes should be timed to the response window of the variable you’re trying to influence, and they should not be repeated until that window has had time to speak.

That is falsifiable. If you adjust less often and the athlete’s trend becomes easier to interpret, the rule worked. If you adjust constantly and keep getting contradictory readouts, the rule failed. But in most physique coaching cases, the failure is not that you waited too long. It’s that you never waited long enough to know.

For coaches, that means a simple discipline:

• don’t cut twice because one morning was flat,
• don’t add twice because one meal looked small,
• don’t rewrite the plan because travel, water, or digestion distorted a single check-in,
• and don’t let AI turn noisy inputs into confident overcorrections.

The best nutrition adjustment is often the one you make once, then leave alone long enough to learn from it.

Sources Used

• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/clients/david_lamartina___members-tlssnsjthkmnhfqcscszce25acz_vhdm_x2_xdlpx_i.json
• raw/_consumed/2026-05-26/troponin_supplements_kb.md
• raw/Justin_TT1.txt