Methods & Data

Reproducible by design, proportionate in practice.

We choose methods for the question and evidence—not for decoration. The working record should allow a competent reviewer to understand what was done, why and with what limitations.

Operating method

Five connected stages.

01

Frame

Decision, users, question, outcomes, scope, assumptions and exclusions.

02

Plan

Protocol, source hierarchy, variables, analysis, ethics, risks and acceptance.

03

Build

Search, screen, extract, clean, analyse and visualise with versioned records.

04

Challenge

Technical review, sensitivity, alternative explanations, disclosure and limitation checks.

05

Deliver

Findings, uncertainty, implications, recommendations, files, handover and correction route.

A scientist examining a specimen in a laboratory in Uganda

Context matters

Methods live inside real systems.

Facilities, data access, staff time, ethics, procurement and decision deadlines affect what is feasible. A strong plan makes these constraints visible rather than pretending they do not exist.

Photo: Bruno Ashimwe Winks · Wikimedia Commons · CC BY-SA 4.0

Quality record

What should be traceable.

Minimum records vary with scope, but these controls guide every assignment.
AreaWorking recordReview question
QuestionDecision, intended users, objectives, outcomes and exclusions.Does the method actually answer the stated question?
SourcesSearches, datasets, provenance, dates, access and permissions.Can another reviewer trace where the evidence came from?
DataDictionary, cleaning log, missingness, transformations and access controls.Were changes justified and preserved?
AnalysisPlan, code, versions, outputs, deviations and sensitivity checks.Are results reproducible and inference proportionate?
InterpretationFindings, inference, limitations, implications and recommendations.Are uncertainty and alternative explanations visible?
ReleaseReview, approval, authorship, disclosure, citation and correction route.May this output be shared, and under what conditions?

Reproducibility

A file is not a workflow.

Reproducibility requires structure around the file: inputs, environments, code, decisions, output versions and a person who can explain them.

01

Preserve inputs

Keep lawful raw records unchanged and separate from working data.

02

Use identifiers

Name files, variables, versions and outputs consistently.

03

Automate carefully

Prefer scripted transformations over undocumented manual changes.

04

Record decisions

Log exclusions, deviations, corrections and reviewer decisions.

05

Separate audiences

Preserve technical records while communicating plainly to decision-makers.

06

Close responsibly

Handover, revoke access and retain or destroy data according to the agreed plan.

Need a method another person can review?

Tell us the decision, intended users, desired output and timing. We will help determine whether the assignment is suitable and what a defensible scope looks like.

Request a scoped consultation