Skip to content
Tutorials Center
My learning

Topics

Lesson 56 of 60 ยท Real-World Projects

Project: Run a Bug Triage Session

Run a bug triage session: sort defects by severity and priority, find duplicates, decide which ones block the release and read a small metrics dashboard.

Finding bugs is only half of the work. A real project collects them faster than it can fix them, so someone has to decide what gets fixed now, what waits, and what is a duplicate of something already known. That decision meeting is called bug triage. Done well it takes 30 minutes a week and keeps a team aimed at the problems that matter. Done badly it becomes an argument about every ticket.

In this project you run a triage session on 13 invented defects from the Bookshop store. You will separate severity from priority, spot duplicates, pick the release blockers and read a small dashboard. The code is a stand-in for the board in a tracker such as Jira: real tools keep the same fields, and the same questions get asked in the meeting. See Writing Excellent Bug Reports and Defect Lifecycle and Quality Metrics for the ideas this project puts to work.

What happens in a triage session

Collect New and unconfirmed defects.
Check Is it clear, reproducible and not a duplicate?
Rate Agree severity and priority.
Decide Fix now, next release, later, or reject.
1 / 2

Severity and priority are different questions

Severity describes the damage: a crash or wrong charge is Critical, a wrong result with a workaround is Major, a cosmetic flaw is Trivial. The tester usually proposes it, because it is a fact about the defect. Priority describes the order of repair, and it depends on things outside the defect itself: how many people are affected, whether there is a workaround, a release date, a contract.

A report export that crashes above 50,000 rows is Critical in severity, yet if only one finance analyst ever sees it and she can export in chunks, its priority may be modest. A typo in the company name on the home page is Trivial in severity but may be High priority the day before a launch. Teams that merge the two ideas into one field lose information in both directions.

Our stand-in turns priority into a small, explicit rule so the meeting stays fast. It adds three numbers: severity (Critical 4, Major 3, Minor 2, Trivial 1), reach (Many 2, Some 1, Few 0) and one extra point when there is no workaround. A score of 6 or more is High priority, 4 or 5 Medium, and below 4 Low. This formula is invented for teaching. Real teams argue about their own rules, but writing the rule down is what makes triage consistent.

JS Sort the defects onto a board

Press Run (or Ctrl + Enter) to see the console output here.

No output.

The crosstab at the bottom shows why the two fields exist. Critical defects land in two different priorities, and the Minor password-reset bug (B-206) gets a Medium priority because many users see it, even though the damage is small. Trivial things sit in the backlog. The board is a starting point for the discussion, not a verdict: a person in the meeting can overrule the formula, but should write down why.

Find the duplicates

Different testers, and users, report the same problem in different words. Duplicates waste developers' time (two people investigate one bug) and distort metrics (two defects counted for one fault). A triage lead scans for them and links the duplicates to one master report instead of deleting them, because every extra report is evidence: B-202 tells you that the failure is not tied to one customer.

Human triage uses search and memory. Our stand-in uses a simple text measure called Jaccard similarity: break each title into stem words, ignore filler words, and divide the number of shared words by the number of different words overall. It is crude, and a tracker's duplicate detector is smarter, but it shows the idea and it makes a good first filter. To keep false alarms down, the code only compares defects in the same module.

JS Detect duplicate candidates

Press Run (or Ctrl + Enter) to see the console output here.

No output.

The detector links two pairs: the two checkout failures (B-201 and B-202) and the two accent problems in search (B-204 and B-205). The other same-module pairs it rejected, such as B-212 against the two crash reports (similarity 0.00), show the value of a threshold. Notice what a machine cannot know. B-212 (the card charged twice) sits in the same module and is a different fault. A human confirms every merge, because a wrongly merged defect disappears from the fix list.

Before merging, also check for the other classic triage rejections: cannot reproduce (ask for steps), works as designed (the requirement says so), missing information, and out of scope. Each is a valid outcome, as long as the reporter gets a reason.

Decide the release blockers

A release blocker is a defect that must be fixed before shipping. Not "important", not "annoying": it must be fixed, because shipping with it would lose money, break the law, damage data or make the product unusable. Teams should keep the blocker list short and visible. When everything is a blocker, nothing is.

In our rule, High priority defects are the candidates. After you merge the duplicates, the list is small and checkable. The stand-in below also builds the dashboard a triage lead would show at the end of the meeting: counts by severity and module, average age of the open defects, and the reopen rate.

JS Blockers and a small dashboard

Press Run (or Ctrl + Enter) to see the console output here.

No output.

Two blockers remain: the checkout crash with ten units, which now has two independent reports, and the double card charge. The double charge is also the defect that has been reopened twice, a warning sign worth a conversation: the fix is probably addressing a symptom, so someone should look for the root cause rather than patch a third time. See root cause analysis in Defect Lifecycle and Quality Metrics.

Checkout and Login tie for the most open defects, but checkout holds both blockers. That matches the idea of defect clustering: problems gather in a few places, so spend extra testing time there.

Run the meeting well

The code gives you numbers, but the session is a human process. A few habits make it work:

  • Keep it short and fixed. Thirty minutes, same time each week, with a tester, a developer and someone who speaks for the product.
  • Prepare. Testers confirm that each new report has steps, expected and actual results and evidence before the meeting, not during it.
  • Decide, do not debug. If the discussion turns into finding the cause, park it and assign an owner.
  • Record the reason. "Won't fix: affects one retired browser" is a decision. "Closed" without a reason is a future argument.
  • Use the same definitions every time. Pin the severity and priority definitions at the top of the board.

Review the dashboard for trends, not snapshots. A rising average age of open defects means decisions are being postponed. A rising reopen rate means fixes are not being verified properly. These signals are more useful than the raw defect count, which mostly reflects how hard people are testing.

Check your understanding

Quick check ยท 5 questions

of 5 answered You scored / 5 ยท perfect!

  1. 1. A rarely used admin page crashes the whole server, but only one employee uses it and a restart fixes it. Which classification is most plausible?

    The crash is technically severe, but the limited reach and an easy workaround can lower its priority. Severity and priority answer different questions.
  2. 2. Why should a duplicate defect be linked to the master report instead of simply deleted?

    A second report shows that more than one user is affected and may contain steps or data that help the fix. Linking keeps that information.
  3. 3. What defines a release blocker?

    A blocker is a defect whose presence makes the release unacceptable. Keeping the list short is what keeps it meaningful.
  4. 4. A defect has been reopened twice after being marked fixed. What is the best next step?

    Repeated reopening suggests the fix treats a symptom or was not verified thoroughly. Root cause analysis and a better retest are the sensible response.
  5. 5. Which is a valid outcome of triage for a report with no steps and no evidence?

    Missing information is a legitimate triage outcome as long as the reporter is told what is needed.

Summary

  • Triage turns a pile of reports into decisions: fix now, next release, later, or reject with a reason.
  • Severity is the technical damage; priority is the business order of repair. Keep them as separate fields.
  • Duplicates are linked to one master report, not deleted, and a person confirms each merge.
  • A release blocker must be fixed before shipping; keep that list short.
  • Read the dashboard for trends: hotspots by module, average age and the reopen rate.

Continue with Practice Exercises with Solutions.