What **interoperability guarantees** apply (which markup/attributes are preserved, translated, or may change)?

[toreini]

From #60, Needs to be defined

Group members: please +1/-1 so we can gauge consensus.

[toreini]

Decision 1: Defer Markdown (not in v1) or not?

Proposal: Defer Markdown to V2. Adding a Markdown format purely for symmetry with SummarizerFormat would be scope creep. Note the false-consistency: Summarizer's format shapes the output formatting of generated text, whereas #60 is about input representation and structure preservation, a different problem. We can revisit Markdown on its own use-case merits if required.

[toreini]

Decision 2: Language detection treated separate than translation?

Proposal: Language detection operates on textual content, decided separately. Detection has no output structure to preserve and markup is noise in the signal. We specify that detection is performed on the text content (markup is not part of the language signal), so developers don't need the innerText dance. This is a low-stakes win and should not be blocked by the harder discussions.

Decision 3: Land the interop guarantees as web-platform-tests?

Proposal: Yes, provide WPTs. The human-language quality is subjective, but structure preservation is objective and testable. We encode DOM-to-DOM's invariant (attributes preserved, tags intact, edge cases) as WPT so "what happens to <span class="count">?" is answered by a passing test, not a thread.