Your Diet Has a Pattern — Here Are the 8 Common Ones

You describe your diet as “mostly healthy, mostly whole foods, generally Mediterranean.”

Your tracker looks at a week of logs and says: 48% of your calories came from NOVA Group 4 products. Your most common lunch was a frozen entree. You had fish once, legumes twice, vegetables irregularly. Your pattern this week is closer to “high-processed” than “mediterranean.”

Which description is right? They both are, in different senses. Your intention and your actual pattern are different things, and the second one is what drives outcomes.

This is a companion to the nutrition intelligence pillar. That piece covers seven dimensions of food quality beyond calories and macros. This piece goes deep on one of them: pattern classification — what the logs actually say, not what you call it.

---

Why Pattern Classification Is More Useful Than Per-Meal Scoring

A tracker that shows you calories and macros for the day is a ledger. A tracker that shows you a pattern label for the week is a summary. The difference matters because human memory is bad at integrating many small data points into a single concept, and that’s exactly what a pattern label does.

A user who sees “today you ate 1,860 calories, 125g protein, 180g carbs, 70g fat” has four numbers to integrate. A user who sees “this week your pattern was high-protein with moderate processing” has a concept — one that connects to years of popular nutrition discourse and has recognizable failure modes and strengths.

The other reason pattern classification matters is that the long-term outcome research in nutrition is increasingly about patterns rather than individual nutrients. “Mediterranean adherence” predicts cardiovascular outcomes in multiple large cohorts (the PREDIMED trial, EPIC cohort, several smaller RCTs). “DASH diet adherence” predicts blood pressure outcomes. “Ultra-processed percentage” predicts a range of cardiometabolic outcomes. None of these are single nutrients — they’re patterns. A tracker that surfaces the pattern label connects your behavior to the outcome literature in a way individual macro tracking does not.

The Eight Canonical Patterns

A classifier needs a finite set of labels. Eight captures most of the distinct patterns that emerge from real user logs:

Mediterranean. Fish a few times a week, vegetables at most meals, legumes regularly, olive oil as default fat, whole grains more than refined, nuts a few times a week, moderate dairy (often yogurt or cheese), wine optionally in moderation. Low-to-moderate red meat, low processed food. This is the pattern most studied for cardiovascular outcomes and the one with the strongest long-term evidence base.

Ketogenic. Very low carbs (typically under 50g per day), moderate protein, high fat from whole-food sources (olive oil, nuts, fish, eggs, fatty dairy, some animal fat). The distinguishing feature is that the composition produces ketogenesis — not just low carbs on a macro sheet but a composition that sustains ketone production. Dirty keto (low carbs but mostly from ultra-processed fat sources) is a real pattern and gets classified separately, usually as high-processed or sometimes “ketogenic with high processing” depending on severity.

High-protein. Protein is emphasized across most meals, often at or above 1.6g per kg of body weight. Carbs and fat are moderate and secondary. Food sources skew toward lean meats, fish, eggs, dairy, and sometimes protein supplements. This pattern is common in strength-training populations and body-composition-focused dieters.

Plant-based. Predominantly or exclusively plants. Includes strict vegans, most vegetarians, and flexitarians who eat animal products rarely. The distinguishing feature is the calorie share from plant sources (typically 80%+) rather than the absolute absence of animal products.

High-processed. NOVA Group 4 foods dominate regardless of macro ratios. This is the pattern most industrialized-country adults actually land in without noticing, because packaged foods are cheap, convenient, and heavily marketed. It’s also the pattern with the strongest adverse outcome associations, which is why surfacing it matters.

Balanced. A mix of food groups in conventional proportions without a specific theme. This is a legitimate label for many people — not everyone follows a named pattern, and “balanced” is descriptive rather than a failure to classify. The outcome research on “balanced” diets depends heavily on what’s inside the balance — a balanced diet of whole foods is fine; a balanced diet with 50% NOVA-4 is the high-processed pattern in a different coat.

Low-carb. Carbs restricted (typically 50 to 150g per day) but not to ketogenic levels. Sometimes called “moderate low-carb” or “Atkins maintenance.” Often used for body-composition purposes rather than metabolic ones.

High-carb. Carbs dominant — often 60% or more of calories — typically in athletic, endurance, or cultural contexts. Traditional Asian and Latin American diets often classify here. Endurance athletes frequently run this pattern during heavy training blocks.

These eight labels are not exhaustive. A meat-heavy carnivore pattern, a raw food pattern, or a very specific cultural pattern (traditional Okinawan, for example) would need more labels to capture precisely. For the vast majority of user logs, eight is enough.

How Classification Actually Works

A classifier has two plausible implementations — rules-based or LLM-based — and they have very different failure modes.

Rules-based classifier. A series of thresholds: if fat > 60% and carbs < 10%, classify ketogenic. If fish frequency > 2x/week and olive oil is default fat and vegetable frequency is high, classify mediterranean. Etc. The advantage is that the logic is inspectable and consistent. The disadvantage is that real user data is messy enough that rules either get very complicated or produce a lot of “balanced” classifications for diets that clearly have a theme.

LLM-based classifier. Take a week of meal descriptions, send them to a language model, ask for the best-matching pattern label. The advantage is that LLMs are good at the kind of fuzzy recognition this problem needs — they can tell that a week with a lot of “avocado toast, tuna salad, grilled chicken, rice bowls” is probably high-protein even if one specific rule doesn’t fire. The disadvantage is that the classification is not fully inspectable and can drift if the model is updated.

A reasonable implementation uses the LLM-based classifier but shows the user which meals contributed most to the classification, so the label has evidence attached. The user can disagree — “that week was high-processed because I was traveling, not because that’s my usual pattern” — and the weekly timeline shows whether it’s a one-off or a trend.

The LLM approach also handles the edge cases. A week that is genuinely a mix of mediterranean and high-protein (because you increased protein while keeping the mediterranean base) gets classified as one with the other flagged as “secondary,” which a rules-based system typically gets wrong by calling it balanced.

Why Weekly, Not Daily

Daily classification is overreactive. A Sunday brunch of pancakes doesn’t make your pattern “high-carb” if your other six days are mediterranean. A single high-processed lunch doesn’t make your week high-processed. The variance at the day level is high enough that labeling daily feels wrong to users — they correctly recognize that one meal or one day doesn’t define them.

Weekly classification smooths the noise. A rolling 7-day view captures the real pattern while staying responsive to actual changes. If you shift from mediterranean to high-protein over two weeks, the weekly label will shift with a week of lag — acceptable because the underlying phenomenon is a pattern change, which itself takes a few weeks to be real.

The eight-week history is the useful visualization. A timeline of weekly labels shows:

• Stability (eight weeks of mediterranean — you have a pattern, it’s stable, biomarkers will tell you if it’s working)
• Drift (four weeks mediterranean → four weeks high-processed — this is a surveillance signal)
• Oscillation (alternating weeks of different patterns — usually a sign of inconsistent eating or inconsistent logging)
• Intentional shifts (deliberate transition from balanced to high-protein for a specific training block — the label confirms the behavior change)

What the Classification Can’t Tell You

The pattern label is a compression of the week’s logs. A few things it inevitably loses:

Whether the pattern is working for you. The classifier tells you what pattern you’re in, not whether that pattern is delivering outcomes. Mediterranean is a good pattern on average; it might still produce poor energy for you specifically if your activity level demands more carbs than mediterranean typically provides. The label is a starting point for asking about outcomes, not an answer to them.

Micronutrient adequacy. Pattern and adequacy are correlated but not identical. A plant-based pattern is often low in B12 without supplementation regardless of overall quality. A ketogenic pattern often runs low in potassium and magnesium. See Tracking 35 Micronutrients for the adequacy layer.

Calories. You can be mediterranean and overeating, or mediterranean and underfueling. Pattern is orthogonal to energy balance. For the underfueling side, see the energy availability guide.

Social and cultural context. A pattern might match your cultural eating norms (traditional high-carb Asian diet, for instance) or conflict with them (vegetarianism in a meat-centric social environment). The label doesn’t capture the context, just the composition.

When Your Pattern Surprises You

The most common useful surprise from pattern classification is the gap between self-description and actual pattern. People routinely describe themselves as eating “clean” or “mostly whole foods” and classify as high-processed once a full week of logs is in. This is not a failure of self-perception — it’s a failure of integration. You remember the whole-food meals clearly; you don’t remember the three coffee-shop snacks and the default cafeteria lunch as well.

The other common surprise is pattern shift around life events. Stress, travel, seasonal changes, and sleep disruption all predictably shift patterns. A tracker that shows you “your pattern this week drifted to high-processed — this is the third week since you mentioned a project deadline” is telling you something you already half-knew and giving you a concrete starting point.

A third common surprise is that “balanced” is a common classification for people who thought they had a clear pattern. If your logs show no dominant dietary theme — meat on some days, plants on others, no consistent emphasis — the classifier will call you balanced. That’s not a criticism. It’s accurate.

How to Use Your Pattern

The pattern label is a question, not an answer. A few question types it supports:

“Is my stated goal actually reflected in my pattern?” If you’re trying to eat a mediterranean diet and the classifier says high-processed, the gap is evidence and can guide specific changes.

“Is the pattern I’m in delivering the outcomes I want?” If you’ve been in a high-protein pattern for six weeks and body composition or training are not responding, the pattern might not be the right fit for your context, or something else is off (total calories, sleep, training plan).

“Is my pattern drifting unintentionally?” A timeline showing a stable mediterranean pattern for six weeks followed by three weeks of high-processed drift is actionable. The drift might be explained (travel, stress) or unexplained (a slow shift in default lunch choices) and worth investigating either way.

“What pattern am I moving toward?” If you’re actively trying to shift from balanced to plant-based, the weekly timeline confirms whether the shift is happening. Intent alone doesn’t move the label; actual food composition does.

Back to the Pillar

Dietary pattern classification is one of seven dimensions the nutrition intelligence pillar covers. The others — NOVA processing, polyphenol diversity, meal-level glycemic response, chrono-nutrition, IARC carcinogen exposure, 35-nutrient tracking, and meal photo analysis — each contribute inputs to the pattern and can be examined on their own. For the siblings most directly related to pattern — since a pattern is partly a summary of processing and macro composition — see NOVA Groups and Polyphenol Diversity. For the cross-cluster conversation on what a pattern does to training outcomes, the adaptive training intelligence guide.

The existing post on this theme is most nutrition trackers count calories — ours understands your diet, which gives the product-angle view.

Omnio’s weekly dietary pattern classifier uses a language model over the week’s logged meal descriptions and writes the resulting label to the time-series store as a first-class metric, so you can see not just today’s macros but the eight-week pattern history on the nutrition page. The classification refreshes weekly on Monday morning, which is when pattern effects actually show up.