Methodology Overview

Root Signal's monitoring workflow is built around a consistent chain from source to publication:

1. Source registry — radio, print, television, and public social/institutional accounts are added to a defined monitoring list.
2. Capture and archive — the original programme, article, post, or document is captured with a timestamp, so what was published can be checked later.
3. Transcription / OCR — audio, video, and image-based content is converted to text.
4. Original-language preservation and translation — content is kept in its original language (English, Swahili, or Luganda) alongside any translation, so nothing is classified only from a translated version.
5. Human coding — a reviewer classifies the relevant segment against a defined framing standard.
6. Quality review — classification is checked before publication.
7. Publication — the reviewed signal becomes part of the public dataset.
8. Versioning and corrections — published records can be corrected; see Verification & Corrections.

AI tools assist with steps 3–5 (transcription, translation, and suggested classification) to make continuous multilingual monitoring practical at scale. They do not replace the human review at steps 5–6.

What this is informed by. Root Signal's approach draws on published, named frameworks rather than an unspecified claim of "international standard": the OHCHR human rights indicators framework for grounding measurement in a defined conceptual basis, the Berkeley Protocol on Digital Open Source Investigations for capture and evidentiary handling principles, and OCHA's data responsibility guidance for safe and ethical data management. We reference these because they are checkable, not as a substitute for publishing our own methodology in full.

Current scope. A fully published, versioned framing codebook and public actor/source registry are in development and not yet live. This page will be updated with direct links as each is published, rather than describing them as available before they are.