How We Work With Evidence
A methodology for turning questions into verifiable, usable evidence.
Research at Dataful begins with a question, not a dataset.
We start by defining what we are trying to understand, what evidence would be needed to answer the question, and what would constitute a sufficiently reliable basis for a finding. Only then do we begin looking for data, documents and other forms of evidence.
Information is rarely evidence on its own. A database may contain thousands of records but still leave the central question unanswered. A government document may establish that something was officially recorded but not that it happened exactly as described. A company disclosure may provide important facts while reflecting only what the company is required or willing to disclose.
Our research therefore focuses on building chains of evidence rather than accumulating sources.
From question to limits
Define what must be established
We begin by defining the question
Every investigation or research project starts by turning a broad subject into questions that can be tested against evidence. Instead of beginning with “What information can we find?”, we ask: What are we trying to establish? What would we need to know to establish it? What evidence should exist if a particular explanation is correct?
This determines what data we collect, which documents we search for, what variables we construct and where we look for corroboration. The question may change as new evidence emerges. When that happens, we document the change rather than forcing the evidence into the assumptions with which the research began.
Map and collect relevant sources
We identify the evidence that can answer it
Depending on the research, this may include government records, administrative data, corporate disclosures, regulatory filings, procurement records, court documents, international databases, technical registries, archived webpages, satellite or geospatial information, social-media data, surveys, interviews and previously published research.
Different sources perform different functions. Some establish identity. Others establish dates, transactions, relationships, quantities, ownership, behaviour or institutional responsibility. Some provide context but cannot independently substantiate a finding. We ask what each source is capable of proving and what it cannot prove.
We establish stable points of reference
Names, organisations, projects and events can appear differently across datasets and documents. Where possible, we identify a stable reference around which evidence can be organised – such as a company registration number, procurement identifier, vessel IMO number, geographic coordinate, contract number, account identifier, date or case number.
We build the evidence across sources
Important findings rarely depend on a single document. The objective is not simply to find several sources saying the same thing. It is to determine whether independent pieces of evidence fit together. Where they do not, the disagreement itself becomes something to investigate.
Turn records into comparable data
Much of the evidence needed for research does not arrive as a clean dataset. It may exist across hundreds of PDFs, webpages, spreadsheets, reports, filings or individual records. We extract the relevant information and convert it into structured data so records can be compared systematically.
We define variables, standardise names and dates, identify duplicates, document exclusions and preserve links to the underlying evidence wherever possible. When analytical categories are created by Dataful rather than supplied directly by the source, we distinguish those derived variables from original source data.
Find patterns, exceptions and relationships
Once the evidence has been structured, analysis begins. Depending on the question, we may examine changes over time, distributions, concentrations, relationships between entities, geographic patterns, networks, financial flows, differences between groups, anomalies or recurring behaviours.
Quantitative analysis helps us identify patterns. It does not automatically explain them. We return repeatedly from the aggregate finding to the underlying records and examine unusual numbers, concentrations or relationships in their original context.
We distinguish evidence from inference
Research often requires inference. Several independently established facts may together support a conclusion that no individual source explicitly states. When we make such an inference, we distinguish what a source directly establishes, what our analysis shows, and what we infer from the combination of those findings.
Test findings and competing explanations
We test alternative explanations
Finding evidence that supports an explanation is only part of the research process. We also look for evidence that could contradict it. We ask whether the observed pattern might result from incomplete data, changes in reporting practices, duplicated records, differences in definitions, selection effects or another plausible explanation.
We verify before we conclude
Verification runs throughout the research. Names are checked against identifiers. Dates are compared across sources. Numerical totals are recalculated. Important records are traced back to their original source. Findings produced from structured datasets are checked against individual records.
For consequential findings, we seek corroboration from evidence independent of the source on which the original finding depends whenever such evidence is available. If two credible sources conflict, we investigate the discrepancy and disclose unresolved contradictions when they materially affect the finding.
State what the evidence cannot establish
A methodology should explain not only how a conclusion was reached but also where the evidence stops. Public records can be incomplete. Databases can omit cases. Documents can remain confidential. Historical webpages can disappear. Corporate disclosures may contain only information required by regulation. Social-media datasets may represent only observable activity rather than the full information environment.
If we cannot establish something, we say so. If the evidence supports an association but not causation, we do not describe it as causal. If available records cover only part of a period or population, we do not generalise them to the whole. If an important document is unavailable, we identify that absence rather than filling the gap with assumption.
We preserve the route back to the evidence
A reader should be able to move from a conclusion to the analysis behind it, from the analysis to the underlying records, and where possible from those records to the original public source. For major investigations, we may publish datasets, sourcebooks, methodological notes, evidence tables or explanations of how the research was conducted.
What we found – and how we know.
The purpose of this process is not to make research appear certain. It is to make the basis of our conclusions visible.
We find and structure data. We connect records across sources. We analyse what they show. We test findings against alternative explanations. We verify important claims independently. And we identify the point beyond which the available evidence does not allow us to go.
That is how Dataful turns questions into usable evidence.