[BUG] Weekly usage meter resets ~4 days early (overnight) despite /usage stating "Resets Sat 5:00 AM" -- Max (20x)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

On Max (20x), the /usage panel states both weekly pools reset on a fixed boundary -- "Resets Sat 5:00 AM". Despite that, the "All models" weekly meter reset overnight Mon Jun 1 -> Tue Jun 2, roughly 4 days before the stated Saturday (Jun 6) reset.

What I saw:

• Stated weekly reset (per /usage and the status line): Saturday Jun 6, 5:00 AM.
• Mon night (Jun 1): "All models" weekly meter ~67% used; countdown "4 days until reset" (consistent with Saturday).
• Tue morning (Jun 2): same meter at 0% -- it reset overnight, ~4 days early.
• The countdown never registered this reset. Instead of jumping to a fresh ~7-day window it kept counting toward the original Saturday boundary. As of today (Wed Jun 3) the popup still shows ~2 days and /usage still says "Resets Sat 5:00 AM" -- the displayed schedule never accounted for the Tuesday reset.
• I had used Claude very little over the weekend, so this was the weekly pool resetting, not a 5-hour session window rolling over.

This isn't a cosmetic countdown glitch: the canonical /usage view itself names a Saturday reset and the meter didn't honor it. Two distinct faults: (1) the weekly meter reset ~4 days early; (2) the displayed schedule never matched the actual reset.

What Should Happen?

• The weekly meter resets only at its stated boundary (Saturday 5:00 AM).
• After any reset, both the /usage panel and the status-line countdown immediately reflect the new ~7-day window, and always agree with the meter.

Error Messages/Logs

Steps to Reproduce

This isn't deterministically reproducible on demand (it's a weekly-window timing/account issue), but here is how it was observed:

1. Run /usage on Max (20x). Note the weekly "All models" pool says "Resets Sat 5:00

AM".

1. Monday night (Jun 1): weekly "All models" usage ~67%, countdown ~4 days.
2. Check again Tuesday morning (Jun 2): the "All models" meter has reset to 0% -- ~4 days before the stated Saturday boundary.
3. Observe the countdown/popup still points at the original Saturday boundary rather than a fresh ~7-day window from the Tuesday reset.

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

I've only signed up for MAX a few days ago, this was my first reset

Claude Code Version

2.1.143

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

I'm using Claude for Windows, right now version 1.9659.4 but it has updated a couple of times since I first started noticing the weekly usage anomaly.

Screenshot (attached below) shows both weekly pools = "Resets Sat 5:00 AM".

<img width="729" height="264" alt="Image" src="https://github.com/user-attachments/assets/b8e1372a-de1a-4037-b593-63746bbc3c8b" />
<img width="1474" height="844" alt="Image" src="https://github.com/user-attachments/assets/d933b584-d02a-4087-af16-63ddb82d5662" />

Not a duplicate of #52921 -- please don't close it as one. #52921 is a weekly "All models" counter resetting on a recurring ~24-hour cycle. Mine is a single reset that fired ~4 days early, on a different day than the fixed "Saturday 5:00 AM" boundary /usage states -- not a fast cycle, and I've only ever observed the one reset. I reference it because they may share a root cause in the weekly-window logic, but the symptom is different.

Also possibly related (different symptom): #54974 (closed as not planned) -- weekly reset day shifting forward each week.

Note: I'm aware the limits are described as rolling windows. That doesn't explain this -- /usage gives a concrete, fixed reset time (Saturday 5:00 AM) and the meter reset days before it, on a different day entirely. The stated schedule and the actual reset simply do not match.