I just finished a proper Taguchi-style orthogonal array self-experiment (L4 design, 3 factors, 4 weeks) to figure out what actually moves the needle on my daily productivity.

(Taguchi arrays) https://www.youtube.com/watch?v=5oULEuOoRd0&t

Basically you structure the experiment such that you can get double the data for a factor by averaging out the influence of other tested factors. Its the best way of getting data when experiments are expensive and time consuming.

Factors tested

• Midday Caffeine: Yes (second matcha ~5-6h after waking) vs No (only 1 morning dose = roughly 1 coffee worth of matcha)
• Morning Walk: Yes (15 min outside walk) vs No (5-min indoor exercise)
• Carnivore Diet: Yes (strict — no bread/rice/seed oils) vs Normal (high-carb Japanese food + occasional fruit)

I tracked sleep hours/quality, productive hours (strictly defined — see below), peak energy/focus, mood, how tired I appear, satisfaction, plus tons of qualitative notes. I kept mild disruptions (dinners, game nights, etc.) in the data instead of marking them as outliers. Ended up with 28 clean days of data.

Productive hours definition: Every morning I start a timer as a kind of ritual. From that moment I am no longer allowed to do any of my usual unproductive habits. I try to focus only on productive work until I eventually give up and stop the timer. I can restart it later in the day, but rarely do. Edge cases are things like listening to important audiobooks while eating out or on outdoor walks. unsure if it should count for as much next time.

Results

Weekly averages for productive hours:

• Week 1 (Caffeine Yes + Walk No + Carnivore Yes) → 4.39 h
• Week 2 (Caffeine No + Walk Yes + Carnivore Yes) → 3.27 h
• Week 3 (Caffeine Yes + Walk Yes + Normal) → 4.11 h
• Week 4 (Caffeine No + Walk No + Normal) → 3.47 h

Main effects on productive hours:

• Midday Caffeine: +1.12 hours (by far the strongest factor)
• Morning Walk: basically neutral / very slightly negative
• Carnivore: +0.16 hours (small positive)

Carnivore also gave a noticeable subjective mental clarity boost (“clearer head, more present, less overstimulated”) and slightly better sleep quality, even if raw productive hours didn’t always reflect it.

Other interesting findings

• Managing electrolytes seems extremely important. When I excluded the 9 days that had clear electrolyte/keto-flu issues (twitches, salt testing, wired-but-tired feeling, etc.), carnivore’s effect on productive hours jumped from +0.16 h to +0.78 h. I will be adding 1 teaspoon of salt a day for the next run. Worth noting a traditional Japanese diet has almost 3 times as much salt as I will be supplementing. Also worth noting, because carnivore was 2 weeks in a row, and non carnivore was 2 weeks in a row, adaption should have been less of an issue than if it were otherwise.
• Post-lunch energy crashes happened on both diets (fatty meals were especially bad). I will try a small breakfast + small lunch + big dinner next time.
• High-carb diet for 2+ weeks → zero notable acne.
• Keeping an open window and maintaining a very low ~600 ppm CO₂ count was very interesting. I think this is something more people should experiment with.
• Salt/electrolytes reliably fixed most keto-flu symptoms (eye twitch, muscle spasms, wired-but-tired feeling).
• Strangely how tired I "appear" in the mirror had only a small effect on productive hours.
• Sleep quality seemed to matter more than sleep quantity, perhaps how much I was sleeping was more indicative that I was sick or electrolyte deficient.

Lessons learned for future experiments

• Need better metrics: a separate Work Quality column + a repeatable way to measure brain fog/overstimulation. Rating work quality may do some of that work.
• Many of my old metrics overlapped too much (satisfaction basically = productive hours).
• Be extremely precise about what counts as “productive hours” (I’m still refining the definition).
• Adding buffer days for weeks with mild disruptive events was a good middle-ground approach (instead of marking them as outliers, I just added 1 more day to the week to average it out).
• Measurement timing drifted as my sleep schedule did — next time I’ll lock in a consistent rating time.
• I may try switching the 1–10 rating system to a comparative one (rating today vs yesterday) and then normalizing the data in the final analysis.

Current winning protocol I’m sticking with

Midday matcha (2 doses) + 5-min indoor exercise only + carnivore (or relaxed carnivore) + my background habits (open windows + low CO₂, morning light, supplements, 6pm lights off).

I’m already planning the next round as an L8 array with more variables that I’m still unsure about (including meal timing and better mental-clarity metrics). that version of the experiment will be shifted more towards carnivore and more caffeine.

I really need a repeatable and high signal metric for measuring brain-fog/"presentness". The most potent one I noticed is walking through crowds, which when off keto produced massive over stimulation and irritation compared to when I was on keto. Problem is it is maybe a bit hard to replicate. I thought about irritating tests like the stroop test and similar ones... Maybe I could play an incredibly irritating game like League of Legends to test it. Actually though maybe something like "getting over it" is not a bad idea.

Until i figure that out, and what new 7 factors to test, I will take a break on the testing for now. I can post the raw data if there is any actual interest, still analyzing the data.