recipe-scale-chemist

Use when scaling a recipe up or down, converting a recipe to a different pan size, or diagnosing why a doubled recipe tastes wrong. Computes non-linear ingredient scaling (salt, chilies, dried spices, leavening, fat, alcohol scale with exponents, not multiplication), pan-area yield factors, and bake/roast-time corrections with the food-science reason for every adjustment.

Overview

Doubling a recipe by doubling every ingredient is the most reliable way to ruin it. Salt, chilies, dried spices, vanilla, leavening and alcohol do not scale linearly with volume — perception and chemistry are sub-linear — while pan geometry changes both yield and bake time. scripts/recipe_scaler.py applies an exponent model (new = old × factor^p) to each ingredient class, handles whole-egg rounding, converts pan shapes (round → square → rectangle), and adjusts bake time for depth and roast time by the mass^(2/3) law.

When to Use

• Scaling any recipe by ×0.25 … ×6 (dinner party, meal prep, halving a bake).
• Moving a recipe to a different pan (9" round → 9×13, squares → rounds).
• Someone asks "why does the doubled version taste saltier / soapy / boozy?"
• Converting a 4-serving recipe for a 12-person pot.

Don't use for: candy/sugar-work stages (temperature targets never scale), pressure-cooker times (pressure dominates), or bread hydration re-balancing (bakers percentages) — those need their own workflow, not exponents.

How It Works

1. Classify each ingredient name against ~12 regex rule classes (salt p=0.80, chilies p=0.75, dried spices p=0.80, fresh herbs p=0.85, vanilla p=0.90, leavening p=0.90, alcohol p=0.90, fat p=0.95, liquids and bulk solids p=1.00).
2. Scale each quantity: new = old × factor^p. Sub-linear classes deliberately scale less aggressively in BOTH directions (a halved recipe keeps proportionally more salt than arithmetic would give).
3. Eggs round to the nearest half with a beat-and-measure note (1 large egg ≈ 50 g ≈ 3 tbsp).
4. Pan conversion computes area ratios (round d → square side = 0.886×d) and derives the scale factor from pan area, not from serving counts.
5. Time corrections: cakes/bars by depth ratio t2 = t1 × (d2/d1)^1.6; roasts by mass t2 = t1 × (m2/m1)^(2/3); oven temperature stays THE SAME.
6. Emit a warnings block: thermometer targets unchanged, "taste at 80%", bloom-reheat notes for spice-heavy dishes.

Using the Script

bash

# Built-in demo: chili ×3, cookies ×0.5, 9x13 cake -> two 9" rounds
python3 scripts/recipe_scaler.py --demo

# Scale a JSON recipe by factor or to a target serving count
python3 scripts/recipe_scaler.py --recipe myrecipe.json --factor 3
python3 scripts/recipe_scaler.py --recipe myrecipe.json --servings 4:12

# Pan math: yield factor and bake-time correction
python3 scripts/recipe_scaler.py --pan-from 9x13x2 --pan-to 9rx9rx1.5 --time 32

Recipe JSON shape: {"title", "servings", "bake_minutes"?, "ingredients": [{"name", "qty", "unit"}...]}.

Worked Example

Beef chili for 4 scaled ×3. Kosher salt 1.5 tsp → 3.3 tsp (not 4.5 — salt perception is sub-linear); cayenne 0.25 tsp → 0.44 tsp (not 0.75 — capsaicin stacks fast); kidney beans 800 g → 2400 g (bulk scales linearly). Roast geometry: a 2 kg brisket to 5 kg → time ×(5/2)^(2/3) ≈ ×1.84, same oven temp, same internal pull temp.

Common Pitfalls

1. Multiplying everything. Salt and heat are perceived sub-linearly; oversalting is unrecoverable, so the model under-scales them on purpose.
2. Scaling leavening linearly for big jumps. Excess soda tastes metallic and soapy; p=0.90 tames it — verify with the p-value note in output.
3. Changing oven temperature when scaling. Temperature sets the chemistry (Maillard, setting of proteins); only TIME moves.
4. Egg fractions. Half eggs are fine when beaten and measured; the script tells you the millilitre equivalent.
5. Pan depth ≠ pan area. Same area at double depth needs longer bake ((d2/d1)^1.6), not more flour.
6. Scaling thermometer targets. 74 °C poultry does not become 148 °C for a double batch — doneness temperatures never scale.

Verification Checklist

• Every output quantity carries its class and exponent (transparency).
• Eggs rounded to halves, never decimals like 1.33 eggs.
• Pan conversion reports area ratio AND the derived yield factor.
• Warnings block present for |factor| ≥ 2.
• Time corrections cite the law used (depth^1.6 / mass^2/3).