Tegy Usage Accounting Parity

End-to-end evidence for Claude-style usage settings, approximate historical backfill, exact provider usage on new turns, and production queue recovery.

Date: 2026-07-01 Production Worker: c8edb12a-dc63-4dfa-b587-47c5e79bad11 Issue: #597, #607 PR: #600, #601, #609

Conclusion

Parity achieved for the Claude.ai usage surface shape: settings modal, usage limits section, current session meter, weekly all-model meter, reset copy, plan badge, and refresh/last-updated affordance.
Approximate backfill is complete for historical SDK result rows. New production turns use exact OpenRouter/provider cost from turn.usage.updated.
The verification found and fixed a production event queue recovery gap. The stale queue backlog went from 790 unprocessed fragments to 0, then a new user-visible production turn persisted exact usage.

Production Proof Video

The video shows the production usage settings surface, a real production chat turn, the assistant answer, and the updated usage settings surface.

Exact Production Delta

Measurement	Before final turn	After final turn	Delta
Current session used	$0.054400	$0.059722	+$0.005322
Weekly all-model used	$0.561900	$0.567222	+$0.005322
Current session exact rows	2	3	+1
Current session exact cost	$0.009510	$0.014832	+$0.005322

Before JSON After JSON

Production Screenshots

Tegy production usage settings before final proof turn — Production usage settings before final proof turn.

Tegy production chat proof reply — Production chat reply: `usage proof final 2026-07-01`.

Tegy production usage settings after final proof turn — Production usage settings after exact ledger movement.

Queue Recovery Evidence

State	Recoverable streams	Unprocessed fragments	Result
Before internal recovery	2	790	Two stale streams had unprocessed fragments.
After internal recovery	0	0	Both streams drained successfully, zero failed stream ids.
After final proof turn	0	0	Queue remained clear after fresh production usage persistence.

Claude.ai Reference

Claude.ai's usage settings evidence was captured from the authenticated Claude session and used as the UI target for Tegy's settings surface.

Claude.ai desktop usage settings reference — Claude.ai desktop: usage limits, current session, weekly all-model limits, last updated and refresh.

Claude.ai mobile usage settings reference — Claude.ai mobile reference for responsive settings layout.

Local Tegy Parity Check

Local screenshots and a short local video were captured before deploy to verify the surface shape against the Claude.ai reference.

Local Tegy usage settings desktop — Local Tegy desktop usage settings.

Local Tegy usage settings mobile — Local Tegy mobile usage settings.

Verification Commands

Command	Result
`npx tsc -b --pretty false`	Passed.
`npx tsx scripts/test-flows.ts`	Passed.
`npm run build`	Passed. Vite emitted only the existing chunk-size warning.
`npx eslint worker/tegy-app.ts`	Passed.
`npm run lint`	Known unrelated baseline failure remains in `scripts/visual-fuzz-report.ts`.