[ael-dev3]

Home now reuses observed card content heights when exact dimensions and style/class fingerprints still match, retaining the measurement fallback and Back geometry. Private WeakMap registration preserves early-paint markup. Leaving a page closes its popovers immediately so nested previews are disposed without waiting for invisible closing animations; same-page dismissal still animates.

Write controls describe their dialog behavior while permission discovery or key approval is pending. A follow confirmed by the first hand-off check keeps its new follower count when the reader was known not to follow beforehand. Previously that fast confirmation was mistaken for an action that already existed.

Validation:

• Required checks pass: formatting, native/WebAssembly/static Clippy, 607 Rust tests and licenses. Tool dependencies are unchanged; earlier tools pass 28/28.
• Latest focused write/auth/early-paint/lifecycle run passes 48 cases without retries. The two permission cases were rechecked after adding settled-state assertions and response draining; both pass. The original full markup/box comparison remains intact, now with permission discovery deliberately held pending.
• Controlled pre-fix tests reproduce both race classes: exact early-paint parity fails at both widths, and a fast first follow confirmation drops the new follower count. The earlier navigation regression also fails against its pre-fix runtime, restoring two stale panels on Back.
• Strict quiet checks pass 70/72 with two existing profile fixmes; the static CI set passes 17/29 with 12 declared skips, all zero retries. Earlier targeted navigation/hover checks pass 63/63.
• Discovery conserves all previous 909 IDs and adds one fast-follow regression: 910 cases. No workflow, retry policy, performance budget or SpacetimeDB change.

The original production post-opening sample was 126.2 ms; its exact cause remains unproven. Bounded per-attempt diagnostics retain sanitized hover/resource/frame/task evidence after the original assertions. The selected card, 600 ms hover, original click-to-post expression, replies assertion and <100 ms budget remain unchanged. The older intermittent Wasm trap also had no retained underlying stack; lifecycle assertions now retain error stacks.

Exact-head CI and normal main deployment/performance checks remain required before final production verification.