Our Bread Database

When we began studying bread, we compiled thousands of different recipes to build a database, expecting that we could distill the data to define what separates one type of bread from another. We gathered over 200 cookbooks—written in Chinese, Danish, English, French, German, Italian, Japanese, and Spanish—focusing on contemporary recipes but also sampling historic ones. The final database contained 4,320 recipes and 6,782,400 million data points, which we hoped would help us empirically categorize different breads, develop insights, and also answer seemingly simple questions like “How similar are different brioche recipes?” and “What distinguishes focaccia from ciabatta and related breads?” and many other musings.

It seemed a reasonable goal, given the limited number of ingredients involved and bread baking’s reputation as a precise science, similar to that for pastry. No one, after all, talks about adding yeast “to taste” in bread—rather, we talk about tenths of grams. We thought the formulas and recipes in professional cookbooks would surely cluster around the specific ingredient proportions that make one bread unique from another. So we analyzed the millions of data points from our hundreds of books, we compared ingredients and techniques, we baked endless loaves from every family of bread, trying to tease out the differences among them.

What we learned from this massive endeavor, to our surprise, was that there are no inviolable truths defining different breads. Even leading bakers disagree on just about every aspect of every type of bread. A cluster of baguette recipes in the database contain similar amounts of water and yeast, but other baguette recipes wildly diverge. Challah is traditionally dairy-free so that it can be eaten with meals containing meat in accord with Jewish dietary laws, yet plenty of recipes exist for challahs enriched with butter or milk. Some recipes deviate so far from others that it seemed strange to even consider them in the same category, as with one brioche that had no eggs (typically a defining ingredient) but did contain the highly unusual additions of yogurt, orange juice, and spelt flour.

We found two main types of variations in our recipes. In one, recipes began with very different ingredients and methods but produced similar results. In the other, different recipes produced very different results, but the loaves were labeled as the same types of bread anyway.

So how could this possible be? Mostly because bread is a far more robust and forgiving substance than we had originally thought. Within reason, almost any recipe will work. Loaves might not look or taste the same, but they’ll all still be bread. Another major reason for the variations is that bakers (especially those who create cookbooks) don’t all agree on what bread attached to a specific name should be, or they prefer different styles of the same type of bread. Or they compromise on some aspects of the bread for other reasons—to make a simpler recipe, for easier dough handling, and so forth. In some cases, traditions get passed down from baker to baker, with many recipes followed automatically due to the belief that the recipe must be made in just the way it was handed down.

We treated these variations as an opportunity. We have more than 120 baguette recipes in our database, and we baked all the recipes that had unusual proportions and then asked questions about each loaf. Could that particular baguette be better? What effect did each variation have on the loaf? We learned a lot through these experiments, and that lack of rules freed us to abandon tradition and build up our own system of baking logic from scratch for Modernist Bread.