Updated 14 October 2026
Change Failure Evidence Pack
Read-only evidence assembly for Jira Cloud and Jira Service Management teams reviewing change failures, correlated incidents, rollback signals and unresolved follow-ups.
Requirements and installation
A Jira Cloud site, permission to install Marketplace apps, an active app licence or trial and access to the project being examined are required. Install the existing Change Failure Evidence Pack listing through Marketplace, select your site and review the requested read:jira-work permission. Open Jira Apps → Change Failure Evidence Pack and complete the normal Atlassian access consent if requested.
Use and recognition rules
Select a project and press Run analysis. Project discovery prioritizes service projects when available. The analyzer reads up to 100 recently updated issues. Change candidates contain change, deployment or release in issue type or summary; incident candidates contain incident, outage or problem in issue type or summary.
A correlated incident is either explicitly linked from a change or created within 72 hours after the change creation time with a shared label or component. Both records must be available in the scanned sample. Rollback signals match rollback, revert or backout in the change summary, labels or components. Open follow-ups are other linked issues whose status does not contain done, closed or resolved.
Outputs and interpretation
Metrics show changes reviewed, failure signals, sample change failure rate, open follow-ups and average evidence score. The table shows up to 50 rows ordered by evidence score. The score starts at 100 and deducts 20 for no incident correlation, 15 for no assignee, 10 for no component, 15 for open follow-ups and 10 for rollback. These are heuristic review signals, not causal proof, compliance certification or a complete project audit. A zero rate on an empty sample does not establish that no changes failed.
CSV
After a populated analysis, Export CSV downloads seven columns corresponding to the visible table: Change, Summary, Correlated incidents, Rollback, Open follow-ups, Evidence score and Review. This version exports only returned rows; an empty result does not generate a CSV file. Treat exported issue content as untrusted text and import spreadsheet columns as text.
Permissions and data
The inspected manifest requests read:jira-work only. Jira requests use the initiating user through api.asUser(); access depends on that user and the site. The inspected resolver performs read operations, declares no external remotes or fetch domains and does not persist Jira content in app storage. Downloaded CSVs remain under the user's control. See the product privacy policy.
Reproducible tests
- Use a test project visible to the signed-in user. Confirm the app opens and projects load.
- Use a Change issue and an Incident issue in the recent 100-issue sample. An explicit issue link from the change should associate the incident. Alternatively, use a shared label/component and incident creation time between the change creation time and 72 hours later.
- For a rollback case, use a test change with rollback in its summary. Confirm Rollback = Yes and a failure signal.
- Link a non-incident follow-up in an open status and confirm its count. Compare scores to the deductions above.
- Export populated results, inspect the downloaded CSV as text and compare its seven columns and keys against the table.
- Run a project with no recognized changes: expect zero table rows, not a proof of compliance. Confirm inaccessible projects/issues are not disclosed.
These are reproduction instructions; publication of this page does not assert that every live scenario was executed during this documentation update.
Troubleshooting and support
For missing records, check user permissions, recognized issue types/summaries, links, creation dates and the recent-sample limit. API errors can reflect permissions or unavailable Jira services. Contact support@quantmatrixintelligence.com or use support instructions. Include app version, project key, timestamp and sanitized reproduction steps; never send credentials or unnecessary customer content.