Multi-Period Meal Optimization

This skill creates meal plans with per-meal constraints, ensuring balanced distribution of calories and nutrients across breakfast, lunch, dinner, and snacks. This prevents issues like 995-calorie "snacks" that occur with post-hoc meal allocation.

Template-Based vs Multi-Period

Consider template-based first (--template --pattern X):

• Produces more realistic meals (1 protein + 1 legume + vegetables per meal)
• Different foods at each meal by design
• Simpler to use: uv run llmn optimize --pattern pescatarian --template --json

Use multi-period when:

• You need exact per-meal calorie/protein control (e.g., "exactly 500-550 kcal at breakfast")
• You want to restrict specific foods to specific meals (e.g., "almonds only as snack")
• You need equi-calorie constraints (e.g., "lunch within 100 kcal of dinner")
• Template-based doesn't meet your specific nutrient requirements

When to Use Multi-Period

Use multi-period optimization when:

• User wants exact calorie limits per meal (template gives approximate)
• User needs protein targets at each meal (e.g., 30g+ at breakfast)
• User wants lunch and dinner to be similar in calories (equi-calorie)
• User wants to restrict certain foods to certain meals (eggs→breakfast)

Quick Start

For basic multi-period optimization with auto-derived meal targets:

uv run llmn optimize --multiperiod --json

This uses default splits: Breakfast 25%, Lunch 35%, Dinner 35%, Snack 5%.

Workflow

Step 1: Gather Daily Targets

First, get the user's overall daily targets:

• Total daily calories (min/max)
• Total daily protein minimum
• Any other daily nutrient constraints

Step 2: Define Per-Meal Targets

Ask the user about meal distribution. Offer these presets:

Balanced (default):

• Breakfast: 25% of calories
• Lunch: 35% of calories
• Dinner: 35% of calories
• Snack: 5% of calories (~100-150 kcal)

Front-loaded (big breakfast):

• Breakfast: 35% of calories
• Lunch: 30% of calories
• Dinner: 30% of calories
• Snack: 5% of calories

Back-loaded (big dinner):

• Breakfast: 20% of calories
• Lunch: 30% of calories
• Dinner: 45% of calories
• Snack: 5% of calories

For a 2000 kcal diet with "Balanced" split:

• Breakfast: 450-550 kcal
• Lunch: 650-750 kcal
• Dinner: 650-750 kcal
• Snack: 50-150 kcal

Step 3: Per-Meal Protein (Optional)

Ask if user wants per-meal protein targets:

• Breakfast protein minimum (suggest 25-30g)
• Lunch protein minimum (suggest 40-50g)
• Dinner protein minimum (suggest 40-50g)
• Snack protein (usually 5-10g)

Step 4: Create Multi-Period Profile

Create a YAML profile at /tmp/multiperiod_profile.yaml:

name: multiperiod_custom
description: "Multi-period optimization with per-meal constraints"

# Daily totals (linking constraints)
calories:
  min: <daily_cal_min>
  max: <daily_cal_max>

nutrients:
  protein:
    min: <daily_protein_min>
  fiber:
    min: 30

# Per-meal structure (triggers multi-period mode)
meals:
  breakfast:
    calories:
      min: <breakfast_cal_min>
      max: <breakfast_cal_max>
    nutrients:
      protein:
        min: <breakfast_protein_min>

  lunch:
    calories:
      min: <lunch_cal_min>
      max: <lunch_cal_max>
    nutrients:
      protein:
        min: <lunch_protein_min>

  dinner:
    calories:
      min: <dinner_cal_min>
      max: <dinner_cal_max>
    nutrients:
      protein:
        min: <dinner_protein_min>

  snack:
    calories:
      min: 50
      max: 200  # Keep snacks small!

include_tags:
  - staple

options:
  max_grams_per_food: 400

Step 5: Add Optional Constraints

Equi-calorie constraint (lunch ≈ dinner):

equicalorie:
  - meals: [lunch, dinner]
    tolerance: 100  # Within 100 kcal of each other

Food-meal affinity (restrict foods to specific meals):

food_meal_rules:
  171287: [breakfast]     # Eggs only at breakfast
  170567: [snack]         # Almonds only as snack
  175167: [lunch, dinner] # Salmon for lunch/dinner only

Step 6: Run Optimization

uv run llmn optimize --file /tmp/multiperiod_profile.yaml --json

Step 7: Present Results

The JSON output has a meals structure:

{
  "data": {
    "meals": [
      {
        "meal_type": "breakfast",
        "foods": [...],
        "total_calories": 480,
        "total_protein": 32
      },
      {
        "meal_type": "snack",
        "foods": [...],
        "total_calories": 145,  // Now properly constrained!
        "total_protein": 5
      }
    ],
    "daily_totals": {...}
  }
}

Present each meal separately with its foods and totals.

Step 8: Handle Infeasibility

Multi-period has more constraints, so infeasibility is more common. Check for:

1. Per-meal minimums exceed daily maximum: "Your per-meal calorie minimums sum to 2500 kcal but daily max is 2000 kcal" → Reduce per-meal minimums

2. Per-meal maximums below daily minimum: "Your per-meal calorie maximums sum to 1500 kcal but daily min is 1800 kcal" → Increase per-meal maximums

3. Protein constraints too tight: "Can't get 50g protein at breakfast with available foods" → Reduce breakfast protein or add high-protein breakfast foods

The optimizer's infeasibility_diagnosis field explains the conflict.

Example Profiles

Weight Loss with Balanced Meals (1800 kcal)

calories:
  min: 1700
  max: 1900

nutrients:
  protein:
    min: 150

meals:
  breakfast:
    calories: {min: 400, max: 500}
    nutrients: {protein: {min: 30}}
  lunch:
    calories: {min: 550, max: 650}
    nutrients: {protein: {min: 45}}
  dinner:
    calories: {min: 550, max: 650}
    nutrients: {protein: {min: 45}}
  snack:
    calories: {min: 50, max: 150}

equicalorie:
  - meals: [lunch, dinner]
    tolerance: 75

High-Protein Muscle Building (2800 kcal)

calories:
  min: 2700
  max: 2900

nutrients:
  protein:
    min: 200

meals:
  breakfast:
    calories: {min: 600, max: 750}
    nutrients: {protein: {min: 50}}
  lunch:
    calories: {min: 800, max: 950}
    nutrients: {protein: {min: 60}}
  dinner:
    calories: {min: 800, max: 950}
    nutrients: {protein: {min: 60}}
  snack:
    calories: {min: 200, max: 350}
    nutrients: {protein: {min: 20}}

Commands Reference

# Quick multi-period with auto-derived targets
uv run llmn optimize --multiperiod --json

# Multi-period with custom profile
uv run llmn optimize --file /tmp/multiperiod_profile.yaml --json

# With verbose output (shows constraint matrices)
uv run llmn optimize --file /tmp/multiperiod_profile.yaml --verbose

# With explicit food IDs (bypass tag filtering)
uv run llmn optimize --multiperiod --foods 175167,171287,172421 --json

Tips

1. Start simple: Use --multiperiod flag first, then customize with a profile
2. Check the math: Per-meal calorie ranges must sum to cover daily range
3. Read the diagnosis: Infeasibility messages are specific and actionable
4. Snacks are small: Keep snack max at 150-200 kcal to avoid the "995 cal snack" problem
5. Equi-calorie helps: Requiring lunch ≈ dinner creates more balanced days