Spark completion still prompts model switch despite remaining spark quota

What issue are you seeing?

I’m blocked by quota flow after a gpt-5.3-codex-spark turn even when Spark usage is still available.
Specifically, when my normal Codex quota is exhausted but Spark still has remaining budget, finishing a Spark turn still surfaces a usage-limit path that asks whether I should switch model instead of continuing on Spark.

What steps can reproduce the bug?

1. Use gpt-5.3-codex-spark in a session while normal (primary) budget is near/at exhaustion.
2. Finish a Spark turn.
3. On next action, the UI/flow prompts to change model / indicates usage block despite Spark remaining quota.

What is the expected behavior?

If gpt-5.3-codex-spark still has available quota, finishing a Spark turn should not force a model change prompt or block continuation on Spark.

Additional information

I found potentially related existing issues:

• #19868 (global quota blocks Spark)
• #23171 (primary 100% blocks other models in running session)