I hope it helps debugging: A Hebrew Wikipedia user complained that she was translating https://en.wikipedia.org/wiki/Jarmo , and when she loaded it from the dashboard, only the first paragraph was loaded, even though she wrote more.

Another example: https://en.wikipedia.org/wiki/Late_life_depression . Some data was saved, but some was lost.

Another user reporting issues about this. In this case the issue was produced when interacting with categories.

And another one: translating "Bluegrass music" from English to Persian, draft=51811.

Just an idea about a possible cause: What if a save action is initiated while the page is still being loaded? This may cause some or all paragraphs to be overwritten by empty data.

Empty data cannot be save because of database table constraints and the validations that prevent empty data saving. Still puzzled with the reason causing data loss.

So, I went to the user created the Bluegrass music translation at wikimania. and he showed me the draft loading correctly!. To make sure, I myself refreshed his Special:CX multiple times and all the time draft loaded, not data was lost. Suspecting spurious network glitch that aborted the draft loading api hit in the specific case?

@santhosh, the article that Asaf was trying to translate to Hebrew and lost is https://en.wikipedia.org/wiki/Brewster_Kahle .

As per our further analysis, it seems to be caused by changes in source article. We are fixing it in T106424

Two more reports, both to Hebrew:

• Armenian genocide
• Hamidian massacres

And also International Alliance of Women.

Ongoing task and will continue till the end of the release

More reports:
https://www.mediawiki.org/wiki/Topic:Spsqro1nmq58h7j6

In this comment there are details of a partial loss of data, and information being recovered in the wrong place. This seems to suggest some issues in the way content is recovered (probably because of heavy changes in the source article).
This could be potentially avoided if we could load the same version the user started translating (either by default or as fallback when the updated version is not compatible).

One more report from the talk page:

https://www.mediawiki.org/wiki/Topic:Sqa0r5zzj199npjp

Change 246464 had a related patch set uploaded (by Santhosh):
Log the save and restore failures to eventlogging system

https://gerrit.wikimedia.org/r/246464

Change 246464 merged by jenkins-bot:
Log the save and restore failures to eventlogging system

https://gerrit.wikimedia.org/r/246464

The simple logging of failures is getting deployed this week. We probably need some more to identify all the causes.

Another report: https://www.mediawiki.org/wiki/Topic:Ss99z5zz77113ikk

https://www.mediawiki.org/wiki/Topic:Ss9twg17zqnwcz0l

https://www.mediawiki.org/w/index.php?title=Topic:Smd55pthmhu4usdb&topic_showPostId=ss56al89q0u7af50#flow-post-ss56al89q0u7af50

yet another report https://www.mediawiki.org/wiki/Topic:Sthgtrqm9nn1fh1o

https://www.mediawiki.org/wiki/Topic:Stzi6ktrgvlm816z

From mid-sprint refinement meeting today:
RB: Do we need more tickets about restoration problems?
ST: Last week I saw one report on the talk page - did not investigate - needs investigation
ST: Best thing is to create tickets for such things on the talk page and add to current sprint
PG: I recently created a ticket which I was not sure had technical issues.
ST: Yeah the one about using older version if alignment fails. We had an issue because we did not save the original source revision, but that has been fixed now.
PG: But does that sound like a good approach we want to take?
ST: There is the tangential issue of stable section ids. There was long discussion when sections should change ids.
ST: We can avoid the requirement of stable ids by using the cxserver id system.
ST: But we also started storing the source sections now – that is the connection.
PG: But we are only storing the sections that user is translating.
ST: We need to get all these things sorted.
ST: If we make the saving and restoring granular, then the scope of losing whole translation is going away.
ST: There is tech debt when we designed and implement parallel corpora. We still save the whole translation in addition to the sections. We want to get rid of saving the whole translation.
ST going over T124399: Migrate the draft translation restore mechanism to use data from cx_corpora table
ST: We can start this (task above) immediately after AbuseFilter work
NL: Should be extra careful to not lose any data

This comment describes a case of translation not being saved after working on it for a while ("a whole day").

@Amire80 / @Arrbee: All subtasks of this high priority Tracking-Neverending task have been closed since 30 months ago and Tracking tasks are deprecated.
Can this task be resolved? If not, what is it open for? Thanks!