From Question to Evidence
A Living Field Guide for Qualitative Research
Nearly a decade ago, we began writing this book. We already had a workflow in mind. It began with the foundations of a research project: a bounded question, explicit concepts, and a record of the assumptions that shaped the inquiry. It then followed evidence through collection, structure, preservation, analysis, and a reproducible path from source to claim.
The information environment was already expanding faster than one researcher could absorb it. Search engines, digital archives, smartphones, and cloud storage made acquisition easier, but they did not make a large and heterogeneous corpus easier to reason about. The workflow exceeded the practical capacity of the available tools. What we were missing was processing power.
The recent arrival of large language models changed that constraint. They can propose concepts, generate search terms, compare excerpts, trace patterns, and expose gaps across a body of material in minutes. Yet power is not judgment. The same systems can invent a reference, erase a minority position, disclose protected material, or present an uncertain inference as settled fact. The method came first. Large language models now make parts of it executable at a scale we could only describe when the project began.
We have revived the project around one claim: a researcher needs an evidence map before the collection becomes too large to reason about. The map links a bounded question to concepts, source families, search decisions, evaluated records, known gaps, and a stopping rule. It does not promise omniscience. It makes the limits of collection visible.
The book treats this record as a scholarly Second Brain. Its purpose is not to store more material. It connects capture, organization, distillation, and expression to a visible chain of evidence. Each stage ends with an artifact that records a consequential choice. Taken together, those artifacts show how a question became a corpus, how the corpus became an analysis, and how the analysis became a claim.
The manuscript is written for two readers. The first is beginning a substantial qualitative project and needs a sequence that can be followed without assuming years of tacit experience. The second has conducted research for years but wants to reconsider established practices after the arrival of new search systems and generative AI. Both readers face the same decisions. They differ in how much explanation, contestation, and adaptation they require.
The full field guide follows a qualitative project from its first question to a documented corpus, an analysis, and a public claim. We are publishing it in reviewed modules so readers can use, test, and challenge each part before the first edition closes. The current revision contains two complete modules. The first follows the path from a question to an evidence map. The second carries the same logic into a literature review. A new synthesis chapter examines how review articles and meta-analyses are becoming open, versioned research objects. It also shows how reusable skills, bounded agents, and versioned project context can support discovery, gap analysis, methodology, and collection without transferring scholarly authority to the system. The first Skills and Agents Lab release turns that argument into an inspectable literature-discovery skill, a read-only scout agent, a context template, synthetic benchmarks, and a public revision route. A second bridge chapter shows how AI can accelerate the construction of event databases while preserving provenance, validation, and causal discipline. Historical Sources as Evidence adds the historian’s chain from archival witness and diplomatic transcription to source-bound observation, Historical Claim, and downstream audit. Writing by Voice, Revising by Ear adds a round-trip method for capturing spoken drafts, verifying recognized text, listening to multilingual prose, and recording accepted revisions. A third complete module will connect the collection plan to a responsibly assembled research corpus. Later modules will address preparation, analysis, disconfirmation, writing, sharing, and maintenance.
The first worked example concerns competing accounts of terrorism. It suits the method because terrorism data do not arrive as neutral observations waiting to be counted. Definitions vary. Collection systems change. Sources disappear. Governments, news organizations, researchers, and database teams record different parts of the same event. The example is familiar to our own work. But the method must also survive evidence that researchers help produce. The revised book therefore adds a composite oral-history study of how tenants experienced and remembered municipal redevelopment. That case will carry interviews, consent records, field notes, administrative documents, researcher memos, silence, refusal, and unequal access through the same sequence of decisions.
AI enters the guide as an assistant whose work must be inspected. It may help generate candidate dimensions. It may suggest neglected source families. It may challenge a search strategy. It may compare the filled and empty cells of a map. It may not decide that a source exists, that a claim is true, or that collection is complete. Those remain scholarly judgments.
This is a field guide because research proceeds through decisions made under constraint. Time runs out. Archives close. Languages exceed the research team’s competence. Files vanish. Ethical obligations place some evidence beyond reach. A useful method does not conceal these conditions. It records them.
That record is the subject of this book.