[BUG] Usage limit suddenly reached sooner on the same subscription plan (Windows)

### What's Wrong?

My usable Claude Code allowance appears to have suddenly decreased: I am reaching the usage limit sooner than before while remaining on the same subscription plan.

To clarify, this report is about hitting the usage cap sooner, **not** about the displayed usage percentage unexpectedly dropping.

Please investigate whether there has been a change or regression in usage accounting, rate-limit enforcement, or the allowance applied to my subscription. I do not yet know the cause.

### What Should Happen?

Usage limits should be applied consistently. If the allowance or the way usage is counted has changed, the change should be clearly explained to users.

### Steps to Reproduce

This is an observation from ongoing use, rather than an isolated minimal reproduction:
1. Use Claude Code under the existing subscription plan.
2. Observe that the usage limit is reached sooner than it was previously.

I do not yet have a controlled comparison with identical workloads or quantified before/after measurements.

### Environment

- Claude Code version, verified with `claude --version`: **2.1.260 (Claude Code)**
- Operating system: **Windows 10 Pro, build 19045**
- Interface: Claude Code through the Claude desktop app
- Exact subscription tier, affected limit window, and model: not specified in this report

### Is this a regression?

The behavior changed from my previous experience, but I have not identified a last working version or an exact onset date.

### Error Messages/Logs

No error logs or usage screenshots are attached.

### Related Report

I searched existing issues and found #92201, which also describes usage limits being consumed faster than before. I have not established that my case has the same cause.

Please advise which usage measurements or diagnostics would help distinguish a reduced allowance from increased consumption or an accounting problem.

No repository file, test, or entry point is named. Start by collecting the exact subscription tier, limit window, model, usage measurements, and a controlled workload comparison, then investigate whether allowance, accounting, or rate-limit enforcement differs from prior behavior. Done means the cause is supported by reproducible evidence and the expected limit behavior is documented or tested.

Rédigé par le modèle d'indexation à partir du texte de l'issue.