Problem

The overview screen has the right raw signals, but the first viewport does not clearly prioritize the two core jobs:

• review historical spend across cost, tokens, and elapsed time
• jump into the runs most likely to explain slow agent tasks

Current screenshot:

Expected direction

• Make spend and time the first visual hierarchy: total cost, total tokens, elapsed time, P95 latency, top agent, top model.
• Add a clear "inspect first" section with the top expensive, top slow, and critical sessions.
• Reduce equal-weight boxes so the screen reads less like a generic metrics dump.
• Keep the shortcut hints, but make the primary action visually stronger.

Validation notes

Captured from a real local run, not demo data.