How to read this playbook

The playbook follows a dataset from start to finish. The map below groups its chapters into the four phases of building a dataset. Select any chapter to go straight to it.

You do not have to read it in order. Pick the route that fits where you are:

• You are new to building datasets. Read the Introduction, then work through the Foundations chapters in order: Project Management, Data Collection, Data Governance, Annotation Design and Data Quality. Each builds on the one before.
• You know your task and want to avoid repeating work. Start with Before You Start. It shows you how to find existing datasets, judge whether they fit, and decide whether to reuse, extend, or build your own.
• You already have raw data and need it annotated. Go to Annotation Design, then Data Quality.
• You are working with speech, images, video or documents. Go to the part for your modality: Speech, Vision or Multimodal. Each explains what changes once you move beyond text.
• Your language has little or no digital text. Read Long-tail language onboarding before anything else. It covers the orthography, community and ownership decisions that come first.
• You are using LLMs to generate or augment data. Read LLM-Assisted Data for the trade-offs and safeguards.
• You are preparing a dataset for release. Read Documentation, Dataset Lifecycle and Legal, consent, and community IP.
• You are offline or on a slow connection. Download the whole playbook as one PDF from the AfriPlaybook menu in the navbar. It is rebuilt with every release.
• You would rather read in another language. Use the language switcher at the top right. Translations are maintained by the community and grow over time.

Throughout the book you will find templates you can adapt, such as consent forms, annotation guidelines and governance checklists, along with worked examples from real African-language projects and links to datasets and tools you can reuse. New terms are defined in the glossary.