What I learned logging 1,000 meals

I built CalorieScan AI starting in 2023. By the end of 2024, I'd logged a thousand meals in my own app. By the end of 2025, more than 2,500. Here's what surprised me.

What I expected to learn

When I started building, I thought logging would teach me to eat less. The pitch in my head was: "you're probably eating 500 calories more than you think; track for a month and the gap will close."

That was 20% true. The more important lessons were elsewhere.

What I actually learned

*1. I wasn't eating too much. I was eating wrong.* My calorie totals were close to my estimate. My protein was 40% lower than I thought. My fiber was about 50% lower. Once I knew the gaps, the meals shifted.

2. The first three days of any "diet" are noise. I dropped 3 lbs every time I started something new and gained back 2.5 every time I "took a break." The water-glycogen swing dominated. Weighing daily without averaging would have driven me crazy.

3. Restaurant meals are 1.7x what I think. I now multiply by 1.7 mentally when I order out. I'm closer than I used to be.

4. The salad thing is real. A "healthy lunch salad" from a chain is consistently in the 900–1,200 calorie range when I log it. A burger and small fries from the same chain is 800. The salad isn't bad food, but the implied healthiness of "salad" is dramatically misleading.

5. My breakfast was the bottleneck. A bowl of cereal had been my default for years. Switching to Greek yogurt + eggs + a piece of fruit added 30g of protein and fixed 70% of my afternoon snacking.

6. I didn't need the app most days. By month four, I was logging only new meals or new restaurant orders. The ones I'd already logged just got reused. My total daily logging time dropped from ~10 minutes to under 2.

7. Sleep was the variable I most under-counted. On days with 6 hours of sleep, my appetite was reliably 200–400 calories higher than on 7.5-hour-sleep days. The cleanest fitness intervention I ever tried was an earlier bedtime.

The thing that didn't work

• Trying to eat the same thing every day. I made it 11 days. I am not Mark Zuckerberg.
• Long fasts. I tried a 36-hour fast once. I felt awful and made it worse the next day with overeating. 16:8 worked. 36 hours did not, for me.
• Aggressive deficits. A 700-calorie deficit lasted 9 days before I quit. A 350-calorie deficit lasted 4 months.

The thing that did work

• Showing up. Logging when I felt like it. Skipping when I didn't. Coming back without drama.

The boring conclusion of three years of self-experimentation is the same conclusion that nearly every diet research review reaches: the diet you actually do beats the diet you don't, and the difference is not in the food choices, it's in the showing up.

What it taught me about building the app

• Make logging fast, or people will quit.
• Don't reward streaks; reward outcomes.
• Surface trends, not days.
• Tell people, gently, when they should log less.
• Don't punish gaps.

That's most of the design philosophy. The product is a thousand small bets that those principles are right. Three years in, I still mostly think they are.

Three years of data taught me to need three minutes of the app a day. That's success.