@fresh-machine-37269 why is the preflight currently not working?

✅ Done! Looked into it — the preflight check itself isn't gone, but its failure feedback is effectively invisible right now, which is why it looks broken: 1. The green check is misleading. The profile editor shows a ✓ as soon as the API key field is merely non-empty — it doesn't mean the key was validated. 2. Real failures only surface as a transient toast. When the save-time configuration check (the actual preflight against the LLM) fails — e.g. an invalid key — the error is reported via a temporary toast that's easy to miss, and nothing persistent stays on screen. 3. Stale results get discarded. If the draft changes while a preflight is in flight, the result is dropped, and late results don't reliably reflect what you're looking at — so users can save an invalid key thinking they just saw a pass. PR #18940 (

feat(settings): retain LLM profile validation failures

) fixes exactly this — it keeps a failed configuration check visible above the editor actions, maps SDK

LLMAuthenticationError

to credential-specific guidance, and discards obsolete preflight results properly. It's currently open and unmerged, stuck at the scope gate: the linked issue #18727 is still labeled

priority:low

without

ready-for-dev

, so it's waiting on a maintainer triage decision rather than being blocked on code. Given the frustration this causes, it may be worth bumping that priority / fast-tracking the merge.