AI Translation Progress Graph

I activated ai translations on my self-hosted Discourse server yesterday. Everything is going well… sort of.

The translation progress graph is either not refreshing or inaccurate. Here is what I am seeing:

This would indicate that I am at 99% for all posts since Feb 23, 2026.

This is not accurate. I have roughly 3,000 posts for this timeframe. Based on the translation logs, it is currently working on posts from about 6 days ago.

So, does anyone know:

  • What is the refresh rate for this graph?
  • The data explorer query for posts waiting to be translated?
  • The data explorer query for posts translated?
  • The data explorer query for posts that was tried to translate but failed do to some error/reason.

Thank you in advance.

We had to cache this page as it was timing out on big sites.

The translation has basically two steps:

  1. Detect post original language
  2. Translate to every other language

At the beginning, your site will mostly be stuck doing step 1, which we initially showed in this translation progress page but have since removed because feedback said it was “too much information”.

That said, this is good feedback. The progress page cache and lack of information of locale detection is making the experience here awful at the beginning of the process.

Last thing, we set the backfill rate very conservatively, you may want to increase it according to your budget.

I would suggest considering a simpler reporting approach.

Based on my understanding, the translation process works in two stages:

  1. Translate eligible topic data.
  2. Translate eligible post data.

From an administrative and reporting perspective, I am less interested in what is currently in flight and more interested in overall progress against the eligible workload. I would prefer to see reporting based on the configured translation eligibility rules.

For example:

Status

Backfill settings are configured to translate all content created after February 23, 2026.

Area Total Eligible Translated Completed
Topics 25,000 540 450 83%
Posts 400,000 3,700 800 22%

Failed Translations

Post ID Reason
34543 Malformed characters on line xxxx

The current graph appears to show operational activity, which is certainly useful. However, what I really want to understand is how much of the eligible work has been completed.

Personally, I am not very interested in completion percentages by language. A topic or post is either translated or it is not. The key question for me is how much of the configured backlog has been processed successfully.

This approach also seems like it would be more database-friendly since it focuses on aggregate counts rather than tracking progress across every language combination.

If language-specific reporting is still valuable, perhaps it could be exposed through a filter. An administrator could select a language and view the same progress table for that language only.

Just some thoughts.

p.s.

What is the current cache timeframe?