6 Stage 3: Map Source Families
6.1 Orientation
Researchers often begin with the source they know how to find. Political scientists search articles and datasets. Historians enter catalogues. Sociologists recruit participants. Journalists search news archives. Expertise creates access. It also creates blind spots.
Source-family mapping asks a prior question: who could have observed, recorded, preserved, or later reconstructed each part of the inquiry?
6.2 Learn
A source family is a group of records produced through a related institutional process. Cabinet minutes form one family. Local newspapers form another. Oral histories, police files, organizational newsletters, court records, satellite images, social media posts, and scholarly databases each carry a production logic.
The family matters because records are not independent merely because they appear in different files. Ten articles may repeat one press release. Two databases may derive from the same news archive. A memoir and an oral-history interview may preserve the same retrospective narrative. Counting files can exaggerate corroboration.
For each dimension in the concept grid, identify prospective source families and the institutions that created them. Record what each family was designed to do, who could enter it, what it tended to omit, where it may be held, and what conditions govern access.
Mapping should include sources that challenge the project’s institutional center. Official records preserve the state well because the state creates records at scale. Marginalized actors may appear through legal proceedings, advocacy collections, private papers, community media, or testimony assembled after the event. Their relative scarcity is not evidence of lesser importance.
6.3 Worked Example
An incident of political violence can leave several traces. Local reporters may describe the event. Police may open a file. Hospitals may record casualties. A government may classify the incident. An organization may claim or deny responsibility. A database team may later code the event from published accounts. Researchers may then download the record and reorganize it for a new analysis.
Each trace answers a different question. A database can standardize events across time. It may be poorly suited to recovering local meaning. A court file may establish procedural facts while reflecting the categories of prosecution. A claim of responsibility may reveal strategic communication without proving operational control.
The official GTD documentation makes its inclusion criteria and data collection rules available to users.1 That transparency permits a researcher to map the system as a source family rather than treat each row as an unmediated event.
6.4 Try It
Take one row from the concept-and-context grid. Write the names of institutions and people who could have recorded it at the time. Then write who might have reconstructed it later. Separate contemporary from retrospective sources.
For each family, record five judgments in prose: what it can observe, why it creates records, who is absent, how records survive, and how access could distort the sample. Repeat the exercise until every important row has more than one plausible evidentiary route or an explicit explanation for why only one exists.
6.5 Guided AI Workflow
AI can generate candidate source families when the researcher supplies a concept, period, place, and actor set. Ask for institutional record types, vernacular sources, retrospective accounts, and material or visual traces. Require the model to state why each family might exist and why it might not. A usable prompt: “For this concept, period, and place, list candidate source families, including vernacular and retrospective ones. For each, state why it might exist, why it might not, and how to verify it.”
Permitted input: The concept, period, place, and actor set, with no sensitive source locations or participant identities.
Do not provide: Restricted collection details, protected repository locations, or participant-identifying material.
Verify: Confirm every named archive, collection, database, or repository through an official catalogue or institutional source. Fabricated repository names are a predictable failure.
Record: Log candidate families accepted and rejected, the verification source, and the model and date. Keep fabricated suggestions as rejected rather than deleting them from the audit trail.
6.6 Integrity Checkpoint
Access conditions are part of the evidence map, but public disclosure can harm people and violate agreements. Confirm before saving:
No sensitive detail exposed: no source location, participant identity, or restricted collection detail entered an external model.
Abstraction where protected: protected operational detail appears only in abstract description.
Access recorded: each family’s access status and governing conditions are noted.
6.7 Save the Artifact
Save source-family-inventory-v0.1.xlsx with one row per family and links to the concept identifiers it may serve. Required fields are family, producing institution, repository, access status, language, temporal coverage, likely dependence on other families, and known silences.
6.8 Advanced Practice
Source families can change over time. A newspaper’s ownership, a ministry’s filing system, or a database’s collection vendor may shift within the study period. Treat a family as stable only after checking the institutions that produced it. Where the process changed, create separate records for each phase and mark the transition as part of the evidence.
See National Consortium for the Study of Terrorism and Responses to Terrorism, “Global Terrorism Database Codebook: Methodology Inclusion Criteria and Variables,” 2024, https://www.start.umd.edu/gtd/downloads/Codebook.pdf.↩︎