Accessibility statement
Making Local TCV usable by more people
The goal is a keyboard-accessible, readable workflow that does not require a mouse or cloud AI account. Accessibility is an ongoing effort, and the current local-inference technology introduces real limitations.
Commitment
The site aims to follow WCAG 2.2 AA principles where practical: perceivable content, keyboard operation, clear language, visible focus, labelled controls, and status messages exposed to assistive technology.
Current features
- Semantic headings, landmarks, labels, and native form controls
- Keyboard-operable navigation, dialogs, buttons, and text fields
- Visible focus styles and sufficient text contrast
- Responsive layouts and text that can be zoomed
- ARIA live regions for model, extraction, ranking, and drafting status
- Text alternatives for meaningful images
Known limitations
- WebGPU and large model downloads may exclude older devices, mobile devices, locked-down computers, or unsupported browsers.
- Long model operations can make progress difficult to predict even when status text is announced.
- Generated prose may be inaccessible in substance — overly complex, inaccurate, or poorly structured — and requires human editing.
- The interface and generated guidance are currently English-only.
- Scanned image PDFs are not supported because the app does not include OCR.
Alternatives
If the local model cannot run, use the guided interview to identify evidence manually, consult the plain-language How it works page, or work directly from the TCV overview and official agency links. The tool is optional; an official TCV can be completed without it.
Accessibility feedback
If you encounter a barrier, contact Sidney Shapiro through LinkedIn. Include the page, browser, assistive technology (if comfortable sharing), and the task you were trying to complete.
Last updated July 25, 2026.