Show pricing coverage for partial Gateway cost estimates #87

Summary

The Gateway summary presents the cost calculated from a small priced subset as if it covered all usage. When most requests lack pricing, the result is a partial sum but visually reads as the total estimated cost.

Live reproduction

The running snapshot reported:

• 176,092,215 total tokens;
• 1,452 requests;
• 1,448 requests missing model pricing;
• only 4 requests contributing to cost;
• displayed estimated cost: approximately $0.05.

This produces an implausible effective rate of about $0.00027 per million tokens. The backend cost_label states that 1,448 requests are missing pricing, but the primary UI amount does not expose coverage.

Expected behavior

A partial estimate must not be presented as a complete total.

Recommended states:

• fully priced: $X.XX estimated;
• partially priced: $X.XX partial estimate · 4/1452 requests priced;
• no priced requests: Cost unavailable;
• missing usage and missing pricing shown as separate coverage dimensions.

Acceptance criteria

• Return explicit priced_requests, unpriced_requests, priced token totals, and coverage percentage.
• Distinguish no estimate, partial estimate, and complete estimate in the UI.
• Never format unknown/unpriced usage as $0.00.
• Show coverage in the summary card and accessible details.
• Cost charts and grouped summaries carry the same semantics.
• Filters and date windows recompute cost and coverage over the same events.
• Add regression tests for 0%, partial, and 100% pricing coverage.

Related

• Refresh OpenAI pricing and support context/service-tier rates #86 updates OpenAI pricing, but this remains necessary because third-party and newly discovered models can always be unpriced.

Relevant code

• src-tauri/src/gateway/telemetry.rs
• frontend/src/components/StackedUsageChartShell.tsx