Claude weekly (all models) reads 100% via OAuth while claude.ai shows 1% — scoped/session windows correct

[LavellJ]

Existing issues

• I searched existing issues and did not find a duplicate.

What happened?

The Claude provider reports the all-models weekly window as exactly 100.0% used, while the claude.ai app's usage page shows 1% for the same window at the same moment. The other windows from the same response are correct (verified against the Claude app):

• Session (5h): 8% — correct
• Weekly scoped "Fable only" (claude-weekly-scoped-fable, from the Claude: surface per-model weekly quota carve-outs (Fable 5) from limits[] — port of steipete/CodexBar#1851 #166 port): 2% — correct
• Weekly all models (secondary): 100.0% — wrong (app says 1%)

The bad reading appeared suddenly (a fetch minutes earlier returned 0.0%) and then stayed pinned at exactly 100.0 across repeated fetches. The suspiciously exact 100.0 suggests a missing/null field being defaulted to exhausted — possibly the legacy seven_day field going null now that Anthropic is migrating window data into the limits[] array (the same migration that motivated #166; the real all-models value should be available from the kind == "weekly_all" entry in limits[]).

CLI output

$ codexbar-cli usage --provider claude --json
[{"cost":null,"provider":"claude","source":"oauth","usage":{"extra_rate_windows":[{"id":"claude-weekly-scoped-fable","title":"Fable only","window":{"is_informational":false,"resets_at":"2026-07-26T22:59:59.671595Z","used_percent":2.0,"window_minutes":10080}}],"login_method":"Claude Max 5x","primary":{"is_informational":false,"reset_description":"Jul 20 at 4:29AM","resets_at":"2026-07-20T04:29:59.671218Z","used_percent":8.0,"window_minutes":300},"secondary":{"is_informational":false,"reset_description":"Jul 26 at 10:59PM","resets_at":"2026-07-26T22:59:59.671246Z","used_percent":100.0,"window_minutes":10080},"updated_at":"2026-07-20T00:01:22.822650800Z"}}]

--source cli produces the same result (falls back to source: "oauth").

Affected area

• Provider-specific behavior
• CLI

Steps to reproduce

1. Claude Max 5x account, OAuth source
2. codexbar-cli usage --provider claude --json
3. Compare secondary.used_percent with the weekly (all models) value on claude.ai's usage settings page

App version

0.43.0 (winget)

Windows version

Windows 11 Home 10.0.26200

Additional context

Related: #166 (Fable carve-out port — working great, thanks!). Happy to test a build; this machine reproduces it consistently.

[Finesssee]

Thanks for the report, I will this ASAP.

[Finesssee]

Status

This was a code bug, not a settings issue on your side.

Root cause: the all-models weekly lane was taken from the legacy seven_day.utilization field. When Anthropic leaves that value stale (often 1.0 → displayed as 100%) while the real weekly total lives in limits[] as kind: "weekly_all", CodexBar showed 100% while claude.ai showed ~1%. Session + Fable-only scoped windows were already correct.

Fix: merged in #213 (main @ 3694f236). Secondary weekly now prefers limits[] weekly_all over seven_day.

What you should do

If you have…	Action
Winget / installer 0.43.0	Wait for the next release. This fix is not in 0.43.0. We are still landing more port work before cutting a new package.
Building from main	Pull latest main, rebuild, refresh Claude (OAuth). Weekly all-models should track claude.ai.

No OAuth re-login or settings change should be required once you are on a build that includes #213.

How to confirm after you get the fix

codexbar-cli usage --provider claude --json

Check usage.secondary.used_percent against the weekly (all models) value on claude.ai. Fable scoped extra window should still look right.

If it still mismatches on a post-#213 build, paste a redacted JSON snippet (no tokens) and we will reopen.