Automated target update tracking via MR labels

Problem statement

Target updates that occur directly in GitLab Merge Requests are not being systematically communicated upstream between GitLab and Argos, potentially resulting in:

  • Changes not being reflected in Translation Memory
  • Updates being overwritten in subsequent localization cycles
  • Manual tracking overhead and potential for missed target updates

Target updates affect multiple downstream processes:

  • AI/MT engines - Training data inconsistencies, potential missed prompting changes
  • Translation Memory - Outdated segments in TM
  • Terminology
  • Human Review workflows - Duplicate work for recurring issues
  • Pipelines - Recurring errors that could have been solved upstream

In line with 'outsource first' of minimizing internal automation to only critical pipeline blockers while establishing clear communication channels for all target updates, we need an automated mechanism to identify and communicate content changes that occur within MRs (Target updates) back to Argos and allow Argos to triage and classify the changes and implement them upstream.

Proposed solution

Automated labeling system

  1. Automatic detection: Argo automatically adds a target-update label (label1) to any MR where translated content has been modified
  2. Visibility in Argo: Labeled MRs are visible in Argo to Argos. Argo is the one place in our workflow that has all the relevant data collected (MR with comments and changes, Phrase project information and project details)
  3. Target update backlog: Since target updates are already merged/published, they can be managed as a prioritized backlog by Argos and labeled completed (label2) in GitLab once implemented.

Benefits

  • Eliminates manual tracking in spreadsheets
  • Automatic detection of changes
  • Direct integration between GitLab and Argos
  • Improved knowledge accumulation
  • Clear audit trail and status of all target updates

Manual prototype

Objective: Validate the approach and measure impact before planning implementation in Argo.

  1. Manually label MRs with target-update (label1) when translations are modified
  2. Document labeling criteria and instructions for Argos
  3. Work with Argos team to establish backlog management process
  4. Create RACI matrix for target update responsibilities
  5. Track metrics:
    • Number of target updates identified (# of label1)
    • Time saved vs. current process (Stakeholder estimate)
    • Number of improvements implemented (# of label2)
    • False positive rate
    • Average time to implement updates (time between label1 and label2)

Deliverables:

  • RACI matrix for QA responsibilities between internal team and Argos
  • Metrics dashboard for tracking automation effectiveness for prototype
Edited by Mika Pehkonen