GitLab Well Architected
Cookbook navigation
Prompts / Measure MR Throughput During Duo Trial

Measure MR Throughput During Duo Trial

AnalyticsMerge-RequestsDuo-TrialReportingAnalyzeAdvanced

Get a directional read on merge request throughput during a GitLab Duo trial by comparing cohort vs non-cohort activity across baseline and trial windows.

For Duo Data Analyst

prompt

You are analyzing merge request activity for <customer> engineering organization to measure the impact of the GitLab Duo.

Scope:
- Top-level group: <root-namespace>
- Query merge requests across all child projects in this group.
- Do not use the duo-evaluation subgroup as the reporting scope if it contains no repositories and is only used for membership management.

Trial cohort usernames:
<paste trial participant usernames here>

Time windows:
  - Baseline: <BASELINE_START> to <BASELINE_END>
  - Trial: <TRIAL_START> to <TRIAL_END>

Instructions:

1. Query all merged MRs in <root-namespace> for the baseline and trial windows. Show: title, author, created date, and merged date.

2. Split results into:
  - Trial cohort = authors in the username list above
  - Non-cohort = everyone else

3. For each bucket and window, calculate:
  - MRs created per week
  - MRs merged per week

4. Produce a comparison table with cohort vs non-cohort baseline/trial weekly averages and % change.

5. Add a short narrative covering:
  - where cohort throughput increased, decreased, or stayed flat
  - how the cohort trend compares with the non-cohort
  - caveats such as small sample size, partial weeks, uneven participation, or missing data

6. Include the GLQL queries used so they can be reused in issues or docs.

Constraints:
  - Do not present this as a full productivity analysis.
  - Do not infer quality or efficiency from MR volume alone.
  - GLQL cannot calculate time-to-merge, review turnaround, comment volume, or lines changed.
  - Those metrics require API-based analysis and are out of scope.

Use the prompt

Provide the cohort and comparable baseline and trial periods. Review which records were included before interpreting the difference; MR volume alone doesn't establish that the trial improved delivery.