Data scientists, researchers, and analysts spend a lot of effort making their work reproducible. But once an analysis leaves the notebook, accessibility often gets lost. A carefully designed visualization uses colors, fonts, or layouts that exclude part of its audience. A report becomes a PDF without semantic structure. A chart becomes an image without a meaningful alt text.

This talk is about the accessibility gap in data science: the space between producing correct outputs and producing outputs that more people can actually use.

We will look at practical examples across common data science artifacts: charts, exported images, generated reports, PDFs, and web documents. The focus will be on small, concrete decisions that can be built into computational workflows: accessible color choices, readable typography, useful alt text, better defaults in plotting code, and report generation pipelines that preserve structure instead of flattening everything at export time.

A common reason accessibility gets skipped is that it feels like manual work: something to fix after the chart is exported, after the PDF is generated, or after the report is published. But many parts of accessibility can be supported by tools, templates, checks, and defaults. Alt text can be generated or assisted from plot metadata. Color and font choices can be encoded in plotting libraries. Report templates can preserve document structure.

The goal of this talk is not to turn every data practitioner into an accessibility specialist, but to show that accessibility can become a first-class citizen in data workflows: something we design for, automate where possible, review, and improve over time, just like reproducibility.

Attendees will leave with a clearer mental model of where accessibility breaks in data workflows, a practical checklist for generated outputs, and examples of how open-source tools can help make charts, reports, and documents more usable by default.