How Bread Lab Tests Claims

Purpose

This document defines the ongoing Bread Lab study program used to develop both The Sourdough Formula (the book) and Bread Lab (the app's recommendation engine).

This is not an academic research program, and it is not intended to produce publishable science. It is a practical, disciplined home-kitchen study program conducted by one baker. The objective is simple:

• Become a better baker.
• Become a better teacher.
• Build a recommendation engine whose advice reflects real baking experience rather than assumptions.

Every study answers a specific question that improves the book, the engine, or both. If a study would change neither, it probably was not worth baking.

Philosophy

This program values useful truth over perfect certainty.

The point is not statistical proof. The point is to calibrate one baker's judgment well enough to teach honestly and to build an engine that gives genuinely useful advice. That means:

• A trend seen consistently enough to help a baker is worth teaching, even without a controlled sample size.
• Recommendations are provisional. They get updated when later evidence points elsewhere.
• A study can be rerun. A conclusion can be reversed. Changing one's mind on better evidence is the method working, not failing.

This is honesty, not tentativeness. The goal is to teach what has been seen consistently, and to remain willing to revise it. The engine inherits this posture directly. Its doctrine already holds that a recommendation defers to the dough on repeated contrary evidence. A study-grade finding that later proves wrong in real use will be caught there. That safety net is why indicator-grade findings are safe to act on: they are provisional inputs to a system that watches its own predictions, not permanent rules.

Methodology

The Bread Lab follows a few simple principles.

Change one thing. Each study focuses on a single primary question and, whenever practical, changes only one significant variable at a time. Absolute isolation is impossible in a home kitchen. Reasonable consistency is the goal.

Predict before baking. Before beginning each study, write down: What do I expect? (Sketch the shape — the curve, the knee, where it fails.) Why do I expect it? How confident am I? (Low / Medium / High.) The objective is not to be correct. It is to discover where intuition agrees with reality and where it does not. Anything not predicted in advance cannot later be claimed as confirmation. Over many studies, this builds a second kind of knowledge: where the baker's instincts are reliable, and where they systematically mislead.

Study curves, not individual loaves. Individual loaves matter less than the trend across a study. The question is rarely "was this loaf better?" It is: What direction is this variable pushing the dough? Does the benefit continue, or level off? Where does the dough begin asking for something different? Where does structure begin to fail?

The engine needs the shape of a response, and especially where it bends and where it dies. Map ladders wide enough to overshoot the useful baking range, because the most valuable data point is often the ugly, inedible loaf that reveals where the curve stops paying. A loaf you would never serve can be the one that tells the engine where to stop scaling its advice.

Every study ends with two questions. Book Impact: What, if anything, changes in the manuscript? Engine Impact: What, if anything, changes in a recommendation or curve?

Provenance

Each study carries a permanent identifier (BL-###) so that, later, the book can cite its sources — "this recommendation comes primarily from BL-002 and BL-004" — and the engine can carry the same lineage internally. When a future study refines an earlier conclusion, the provenance chain is traceable. This is nearly free to establish now and almost impossible to reconstruct later.

The Study Record

Every BL study is recorded in this structure. The completed record, with the baker's own observations, becomes the permanent unit of knowledge.

Study ID: BL-### · Pantry version · Question: the single question this study answers.

Prediction (before bake): expected shape / result; reasoning; confidence (Low / Medium / High).

Actual Results: hydration, dough feel, mixing, gluten development, bulk behavior, proof behavior, oven spring, scoring, crumb, flavor, overall handling, surprises.

Conclusions: What did I learn? What assumptions were wrong? What surprised me? Would I change the protocol?

Book Impact: does anything in the manuscript need changing? Engine Impact: does any recommendation or curve change? Confidence: Low / Medium / High. Next recommended study.

The Unit of Knowledge

The completed BL study — not the individual loaf — is the unit of knowledge. A finished study, with the baker's own bench observations, is what advances the project. That is the point of the structure: durable, traceable, provenance-bearing knowledge that compounds.

The Bread Lab is never finished. It becomes smarter every time a better question is asked.