Inspector: study familiar live-stream timeline controls after follow semantics land

[kevinswiber]

Context

Land the focused follow-interaction correction in #517 first. Once that behavior is stable on main, take a deliberate pass over familiar live-stream and scrollback interaction patterns before redesigning the surrounding timeline controls.

The current inspector exposes follow, newest first / oldest first, and start / end across different chrome regions. The labels and grouping do not yet communicate one coherent model of stream ingestion, read position, sort order, and boundary navigation.

Questions to answer

• What do developers already expect from tail -f, terminal scrollback, log viewers, browser developer tools, observability consoles, and other live event feeds?
• Which products preserve viewport position while buffering arrivals, and how do they surface or clear the unseen count?
• Should follow mean ingestion, viewport pinning, or both? If both states matter, how are they named and represented?
• How should the live edge behave under newest-first and oldest-first ordering?
• Are start / end understandable when sort order reverses, or should the actions be named latest / oldest?
• Where should follow, ordering, boundary navigation, and new-event catch-up live so they read as one timeline-local system?
• What keyboard, screen-reader, narrow-layout, and reduced-motion expectations should the design preserve?

Deliverable

Produce a short evidence-backed interaction recommendation for Pointbreak Review, including:

• representative examples with links and captured observations;
• a comparison of the relevant behaviors and tradeoffs;
• the proposed state model, labels, grouping, and keyboard behavior;
• a recommendation for ascending-order catch-up and the disposition of Inspector: surface new-event catch-up in ascending timelines #513;
• focused follow-up implementation issues rather than one broad UI rewrite.

Sequencing

• Blocked on Inspector: keep timeline follow enabled during scrollback and surface unseen events in-list #517 landing and receiving a short period of real use on main.
• Do not reopen the core follow semantics while conducting the control-layout study unless usage exposes a concrete defect.
• Treat Inspector: surface new-event catch-up in ascending timelines #513 as an input to the study, not as an independent assumption that continuous follow must remain descending-only.