Docs/diagnostics request: expose effective context and quota semantics for GPT-6 Astra on ChatGPT OAuth Codex path

Request

Please document or expose diagnostic metadata for the effective context limit, quota behavior, and terminal error semantics when GPT-6 Astra is accessed through the ChatGPT OAuth/Codex path.

Observed bounded validation

Using an OpenAI-compatible proxy backed by the ChatGPT OAuth/Codex path, both requests completed successfully with a short output cap:

• 260,307 actual prompt tokens: HTTP 200, non-empty output, finish_reason: stop
• 300,307 actual prompt tokens: HTTP 200, non-empty output, finish_reason: stop

This shows the path accepts more than 240k input tokens. It does not establish the effective maximum, quota accounting behavior, or what a context/quota rejection should look like.

Why this matters

The public model page advertises a 1.05M context window, but users of the ChatGPT OAuth/Codex surface cannot safely infer that the same effective limit, entitlement, or rate/quota policy applies. Testing near the limit is costly and can consume shared quota.

Requested clarification

1. Is there a supported way to discover an account/path-specific effective context limit for Astra?
2. Are usage, quota/reset, and context-limit errors surfaced in stable machine-readable fields for Codex-backed requests?
3. If a streamed response is ended upstream before normal terminal completion, what retry/resume/background workflow is supported?

No credentials, prompts, endpoint URLs, account IDs, or response bodies are included. This is a documentation/diagnostic request, not a claim that the two successful requests above failed.

No file or test is named. Start by locating documentation and diagnostic handling for ChatGPT OAuth/Codex requests, then determine whether supported metadata exists for effective context limits, quota and reset errors, and streamed termination; done means documenting stable fields and supported retry or resume behavior.