Repeat of sep 9th codex 5hr window gone in 2 requests 20 mins

What version of Codex CLI is running?

0.154.0

What subscription do you have?

Plus

Which model were you using?

Gpt-6

What platform is your computer?

Linux

What terminal emulator and version are you using (if applicable)?

No response

Codex doctor report

What issue are you seeing?

Similar to the events on the 9th September my 5hr window is at 0% around the end of my second task request both run within 20 minutes. So not even 2 goes in 20 mins and I'm empty.
I checked at the start of my session /status : 1:24am expires/renews at 6:24am - 100℅ allowance.

One thing I will note is I was using gpt-6-astra high all day yesterday and just before this planned task, I switched to medium model as it is working from an orchestrator plan file!

HOW can switching down to medium use more, this must be a glitch or error?

Either openai are reducing usage allowance limit daily bit by bit or there's an error like the other day, because yesterday I did similar work and had many many runs without coming close to the limit and edited a lot more code than tonights debacle.

Looking at git the last commit was less than 100 lines of code easily. And the current interrupted run wasn't much more, mostly a line or two changed per file, around 15 files + 1 created file of around 30 lines. Clearly not a lot of code for the usage claimed to have used.
Checking /usage shows a very dim square for today, especially compared to the almost white ones over the last month.

There is clearly something not right, I use codex daily and this was not a normal sesh!

What steps can reproduce the bug?

Unknown, maybe ask gtp6astra high to build a plan file for a multi task goal, then switch to medium and ask it to start the task list on stage 1...

What is the expected behavior?

My 5hr usage not hitting zero after 20 minutes work when yesterday I was closer to 4hrs work before hitting usage limit.

Additional information

No response

Start by tracing the CLI entry points behind /status and /usage, using the reported Codex CLI version, Linux platform, model switch, and two-request timing as investigation details. Compare the five-hour allowance shown at session start with the usage recorded after the two requests; done means identifying a reproducible accounting cause or documenting why the reported usage is expected.