"(migrated)" duplicate fields after a Server or Data Center move: merge or keep
Also published on Atlassian Community
In this guide
After a move from Server or Data Center, the Cloud field list often has pairs: Customer, and Customer (migrated). Both have values, both are on screens, and with 700 fields per space now enforced, each pair counts twice.
LimitRadar is a limits and cleanup app for Jira Cloud from Radrly. This guide is a pair by pair walk: why the pair exists, three questions that decide merge or keep, the native merge, and then how LimitRadar handles the same pairs.
Why the pair exists
Per Atlassian's article on JCMA duplicates and tracking entities, when the migration assistant finds an entity in Cloud with the same name and has no record of migrating it, it adds "(migrated)" to the new one so no data is lost. Repeat runs from several servers, a test server, or Site Import followed by JCMA create numbered copies such as (migrated 2). For custom fields, a different data type, a different description, even a minor one, or different contexts are enough to stop a match.
So a pair is a guess Jira refused to make. Your job is to make it with evidence.
Question 1: are they really the same field?
Compare the type, the options of select lists and the contexts. A single select with options Gold, Silver and Bronze is not the same field as a text field called Customer tier, whatever the names say. Different types mean keep both, or rebuild one by hand.
Question 2: which one is in use now?
Check which field the Cloud screens, filters, boards, dashboards and Automation rules use since the migration. That one is the target. The other is the source whose values you want to keep.
Question 3: do the values collide?
Search for work items where both fields have a value, using the two field IDs from the custom fields list:
cf[10050] is not EMPTY AND cf[10231] is not EMPTY
No results, and the merge is a straight copy. Results with equal values are fine. Results with different values need a decision per work item, and no tool should make it for you.
The native merge
Atlassian's article on merging data from two custom fields gives two methods. A CSV round trip: export work items, move the source column into the target column, import. It overwrites what is already in the target, so it is only safe when question 3 found nothing. Or a scheduled Automation rule with the conditions source is not empty and target is empty, then an Edit work item action that sets the target by copying it from the source. That one never overwrites.
Afterwards delete the (migrated) field. Per Delete or restore a field, it stays restorable for 60 days.
If the migration left a space at 90% or more of a limit, Data limits and guardrails gives a three month exemption for that entity. That is the window for this cleanup.
The same pairs in LimitRadar
The Migrated fields screen lists fields whose names end with "(migrated)", each paired with the original field of the same type. Each scan samples up to 24 pairs, reading two pages of 50 work items per pair, and labels each pair:
| Label | Meaning on the sampled work items |
|---|---|
| Conflict | At least one work item has different values |
| Review | The values overlap and are equal |
| Safe to merge | The target is empty wherever the source has a value |
A merge runs from a plan, only between fields of the same type with compatible options, and only fills empty targets. Conflicting values are never copied. One run reads up to four pages of work items and copies up to 40 values, so a large pair takes several runs. The source field goes to the Jira trash only when a run finished its pages and found no conflict.
A merge cannot be undone in LimitRadar, and the Actions log says No undo. Before the run you need write mode, a stored simulation and the typed plan name plus MERGE. LimitRadar writes a JSON backup immediately before it starts. Forge apps cannot read Automation rules, so (migrated) fields always carry Automation has not been checked; check your rules yourself before the source goes to trash.
Enter your migration completion date in Settings and the Overview shows when the exemption ends and which scopes are at or above 90%. Atlassian sources state the window differently, so you choose three or six months.
The sample is a sample. A pair that was not sampled, or a work item outside those pages, stays at Review. Migrated field merges are part of Standard; threshold alerts and the Rovo agent are (Advanced).
Quick answers
Why does Jira Cloud have custom fields ending in (migrated)?
When the Jira Cloud Migration Assistant finds an entity in Cloud with the same name and has no record of migrating it, it adds "(migrated)" to the new one so no data is lost. Repeat runs from several servers, a test server, or Site Import followed by JCMA create numbered copies such as (migrated 2). For custom fields, a different data type, a different description or different contexts are enough to stop a match.
How do you decide whether to merge a (migrated) field or keep both?
Ask three questions. Are they really the same field (type, select list options and contexts)? Different types mean keep both, or rebuild one by hand. Which one do the Cloud screens, filters, boards, dashboards and Automation rules use since the migration? That one is the target. Do the values collide? Search for work items where both fields have a value; different values need a decision per work item.
How do you merge two custom fields natively without overwriting values?
Atlassian describes two methods. A CSV round trip overwrites what is already in the target, so it is only safe when no work item has values in both fields. A scheduled Automation rule with the conditions source is not empty and target is empty, then an Edit work item action that copies the source into the target, never overwrites. Afterwards delete the (migrated) field; it stays restorable for 60 days.
Can a LimitRadar merge of (migrated) fields be undone?
No. A merge cannot be undone in LimitRadar, and the Actions log says No undo. Before the run you need write mode, a stored simulation and the typed plan name plus MERGE, and LimitRadar writes a JSON backup immediately before it starts. A merge only fills empty targets, conflicting values are never copied, and the source field goes to the Jira trash only when a run finished its pages and found no conflict.