Skip to content

Latest commit

 

History

History
50 lines (28 loc) · 3.55 KB

File metadata and controls

50 lines (28 loc) · 3.55 KB

Our research approach

Back to Fajrix Research

This page describes the publication expectations we intend to apply to research shared here. It is a public overview; the method used for a particular study belongs with that study. It does not certify any existing article as having completed a particular review process.

Start with a defined question

State the question, intended audience, scope and relevant time period. Explain what is being measured or examined and why the chosen method is suitable. Keep exploratory work distinguishable from a test of a previously specified proposition.

Make evidence traceable

Prefer original sources where they are suitable. Record the source title, publisher, date, relevant location and any definitions needed to interpret it. For calculations, show the inputs and transformations that connect a source to the reported result.

Separate source observations from estimates, assumptions, forecasts and Fajrix's interpretation. If several publications repeat the same underlying evidence, make that relationship clear.

Compare like with like

Check definitions, units, time periods, population coverage and country attribution before drawing comparisons. Explain material differences that cannot be resolved. Missing information should remain visible as missing information.

For AI evaluations, specify the task, inputs, scoring method, baseline and relevant configuration. For national comparisons, distinguish measures of spending, technical capability, adoption and value capture.

Explain uncertainty and limitations

Identify what the evidence does not establish. Where relevant, examine sensitivity to assumptions, report failure cases and explain how results could change under different conditions. Distinguish an observed relationship from a supported causal explanation.

For synthetic data or controlled examples, explain their limited coverage of real-world conditions. A reproducible calculation can still depend on uncertain or incomplete inputs.

Make the work inspectable

The supporting material should fit the type of publication:

Publication type Useful supporting material
Quantitative analysis Source register, definitions, calculations or scripts, assumptions and expected outputs
AI evaluation Task specification, permitted test inputs, scoring logic, configuration, results and failure cases
Conceptual framework Definitions, reasoning, source basis, worked examples where useful, and boundaries of application
Policy or strategic commentary Dated sources, explicit assumptions, interpretation and alternative explanations

Where sharing is restricted, explain the limitation and provide source links or access instructions where possible. State reuse terms for released code, datasets and other materials as appropriate to each item, including any third-party restrictions.

Review and correct

Check whether the conclusions follow from the evidence and whether a reader can follow the reasoning. Describe the review actually performed; use the term independent review only when an independent reviewer was involved.

Give substantive releases an identifiable version or date. Record corrections that affect findings, retain a clear change history and explain whether a revision changes a conclusion.

Keep research status clear

Label proposed work, exploratory findings and released companions accurately. A research question is not a result, and a reference example is not a statement of product readiness. Give each companion its own scope so readers can judge what it contributes.