SDL accessibility events: should SDL synthesize focus changes, or should the app map them?

Context

I'm building an SDL2-based UI layer that must work with screen readers on Windows, macOS, and X11. SDL_PollEvent gives me raw input and window events, but no accessibility metadata, so I currently translate SDL_WINDOWEVENT_FOCUS_GAINED and my own widget focus model into platform accessibility API calls (UI Automation, NSAccessibility, AT-SPI) myself.

The decision

SDL 2.0.9 added SDL_HINT_ACCESSIBILITY_FOCUS for focus tracking, but its behavior appears to differ per platform, and I'm unsure how much of the focus-change story SDL is expected to own. I see two documented approaches:

• Rely on SDL's hint-driven focus tracking and treat window focus events as the single source of truth, mapping them one-to-one to accessibility focus events.
• Keep SDL purely as an input source and synthesize all accessibility focus/role announcements in the application layer, per platform.

A complicating constraint: on Windows with high-DPI scaling, event coordinates can be compressed or expanded, so any hit-testing for focus rectangles seems to require SDL_GetWindowDisplayDPI before computing accessible bounds.

Questions

1. Is SDL_HINT_ACCESSIBILITY_FOCUS intended to be sufficient for cross-platform focus announcement, or is it only a partial helper that still requires per-platform app code?
2. Where should DPI correction happen so accessible bounds match what SDL reports in event coordinates?
3. Is there a documented position on SDL ever synthesizing accessibility events itself, or is that explicitly out of scope?