Arguments for data quality

There's no reason to engage with this dataset unless you're convinced it is high enough quality to distinguish signal from noise or systematic error. Below are factors affecting data quality.

• Nearly zero fast food and minimal restaurant consumption (1 meal/week, usually the same meal at one of only maybe 3 restaurants).
• Full time food scale use with increasing attention paid to high caloric density foods. Soy sauce may be added thoughtlessly, but butter, oil, and peanut butter never are. A food scale is always used. I travel with one.
• Binges recorded. Virtually no one uses calorie counting apps when they eat a whole pizza. I do. Every single time. Zero days are missing in 12 years.
• Nearly supernatural taste for calories resulting from years of allocating them to hit 1200 or 1500 calorie targets. Recognizing a food that offered more hours of satiety per calorie than another was such a major win that I developed an adversarial model of preferred foods. When my partner offers a calorie-counted homemade snack and claims it's 250, upon eating it I may counter no, this tastes like 325, and upon inspecting the calculation, find 2 sticks of butter entered as 2 tbsp or flour measured in cups vs. grams. I do this by imagining whether 4 or 6 of this would satisfy me for the day, and if it seems over-unity (too delicious: I'd rather eat 6 of these than anything else to reach 1500) then I know the calculation is probably off. This is also how I estimate restaurant meals when data is unavailable.
• Corroboration. I have location history, frequent time/geotagged photos, and electronic receipts, so if your model asserts there is some mismatch, I can likely help you track it down or amend my data.

Arguments for model subject

Ideally, if you could collect 10+ years of data, you want a subject with the fewest low-hanging fruits. You don't want easily-solved causes of obesity like massive cola or fast food intake or a high-stress life in a food desert or a profession with exposure to industrial contaminants or taking a variety of medicines.

• Actual capture of significant weight loss and regain with many plateaus.
• Leisurely, low-stress, retiree-like life. I do not go into an office. My weekends aren't much different from weekdays. I haven't been through any very high stress situations, no kids, pets, etc. I don't "eat my feelings." I am just hungrier than most people and delicious foods seem to give me significantly more joy than most people.
• Understanding of nutrition labels. I have a good understanding of nutrition and food labels and am mathematically inclined. I know if the label says fibre, carbs are already subtracted. I instantly spot erroneous labels, like ghee having a serving size of 1oz and claiming only 14g of fat.
• No consumption of meat potentially simplifying sources of contaminants.
• No workplace exposure also removes a variety of workplace contaminants. Reverse-osmosis water at home.
• Willingness to test your new ideas for diets as long as they are meatless, non-catabolic, and do not leave me extremely hungry without corresponding results. I do not have a monetary budget for food. The dollar value I place on each pound my set point is reliably pushed down is in excess of 4 figures. I am resilient to high/low fat/carb/protein, fasting, exercise, and weight lifting.
• Willingness to collaborate. You've exported the most common 200 foods I've eaten that account for 80% of intake and need me to go through and validate their MyFitnessPal database entry or export my own custom entries in order to get precise micronutrients for each one? No problem.
• Full-time access to medical-grade instrumentation like a Cosmed Fitmate indirect calorimeter that I can use to monitor changes in RMR.

Arguments against quality

• You can assume an undercount by a consistent 5% due to snacking. I allocate about 100 calories a day to snacks that I don't count. Maybe a square of dark chocolate, three pecan or walnut halves, a cheese straw. This is a source of error, but it's consistent and proportional. On days with 4000 calories like some club-hopping birthday binge, the error scales up accordingly to accommodate cocktails with unknown proportions or a forgotten garlic knot during a night of drinking.
• Increased snacking to maintain momentum. When I had a run of 30+ days of under 1600 calories, the pressure to not screw up increases, and some of those days may have 200 calories of uncommitted snacks where I nibbled on a bunch of small things to deal with ravenous hunger. The longer the run of days, the more likely up to one per week is tainted by an additional 15% unwritten snacking.
• I am still human and can forget to enter something or fail to properly account for some extra oil or butter tossed into a restaurant dish.