[prakersh]

Summary

• Since Menubar never blank, Cursor string numbers (#151), Codex reset credits (#152) #153, 7d and 30d history arrives sampled to about 200 points. Their normal spacing (about 50 minutes at 7d, 3.6 hours at 30d) exceeded the poll-based gap threshold, so every segment was drawn as a faded dashed gap with point dots, for every provider.
• The gap threshold is now the larger of the poll-based value and three times the series' median spacing. Sampled ranges draw as continuous lines, and a real outage (for example a day without data) is still shown as a gap.
• Metadata fields prefixed with an underscore (Anthropic history's _source) are no longer plotted as a series.
• Per-model Anthropic weekly buckets are labelled by name in chart legends ("Weekly Fable" instead of seven_day_scoped_fable).

Test plan

• New node-driven test: evenly sampled 7d/30d series have no gap segments and no interior dots, a day-long outage is still one gap, _source is not a series, and the Fable label resolves
• Related web tests (history, chart, peak, window) pass under race
• Visual check of the Anthropic 7d and 30d charts

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!