Problem

When the AI provider rejects requests during AI pre-translation or TQE (for example because the rate limit of the provider account is reached, HTTP 429), every rejected request is written to the task log as a technical error without its own event code (E9999). A run with many rejected requests fills the task log with such errors.

The entries do not say that the provider rejected the requests or what can be done about it. Users therefore assume a translate5 error, although the cause is the provider account (rate limit, quota, API key) or the language resource settings.

Solution

• Recognize the typical provider-side failures: rate limit reached, quota used up, access denied (API key), provider unavailable, connection problem.
• Add a dedicated event code for each of them.
• Do not write the technical error (E9999) to the task log anymore. The details of each rejected request stay available at level "info".
• At the end of the run, write one warning per failure type with the number of affected segments, a clear statement that the provider rejected the requests, and a hint what to do (for example: lower the parallel requests of the language resource or raise the limit at the provider).
• Same behaviour for AI pre-translation and TQE.
• No additional retries or waiting on a rate limit.