Bug Triage: A Weekly Process That Keeps the Backlog Honest
By Sebastiaan Jansen · · 8 min read
Bug triage is a fixed weekly review of every new or stuck bug. For each one you ask whether it's real, how bad it is, who owns it, and whether it stays. A good triage process keeps duplicates, vague reports and abandoned items out of the queue, so Monday starts with a backlog you can trust.
What bug triage is for
A tracker records reports. Bug triage decides what they mean. Skip it and the list keeps growing, and priority gets argued mid-sprint, usually after a customer writes in a second time.
The meeting has three jobs:
- Confirm whether it's a defect, a duplicate, or something else (a request, a question, a config mistake).
- Set severity, and a priority that matches how you ship rather than how loud the report was.
- Name an owner, a next stage, or an explicit "not now," with a reason someone else can read later.
Reproducing, fixing, and estimating happen on the ticket after the meeting. If you debug in the room, the rest of the queue waits a week. By Thursday, support and on-call still need one place that says "this is next" and "this is parked."
A weekly meeting that finishes
Hold it once a week, same time, for 45 minutes. Add a second slot only after a release, while regressions are still coming in. Daily reviews just replay the same five bugs.
Who is in the room
- A lead who keeps time and writes decisions: an engineering manager, a senior engineer, or a support lead who knows the product.
- One engineer who can say "we can reproduce this" or "this is working as designed."
- One person for users: support, customer success, or the founder on a small team.
Anyone else joins for their ticket and then leaves. Past five people, it turns into a status review.
The agenda
Work from the top of the queue, not by who filed the bug.
| Minutes | Block | Done when |
|---|---|---|
| 0-5 | Queue check | New bugs since last week are listed. Carry-overs are flagged. |
| 5-30 | Decide each bug | Repro call, severity, priority, owner, and stage are set. Duplicates are linked and closed. |
| 30-40 | Stale review | Anything untouched for 30 days is scheduled or closed as "will not fix." |
| 40-45 | Publish | The lead posts decisions where support and engineering will see them. |
Whatever is still open at minute 30 waits. Next week starts with the oldest unreviewed bugs, plus a note that the queue outgrew the slot. The fix is a second slot or cleaner intake. Do not stretch the meeting to 90 minutes.
What "decided" means
Every bug leaves in one of five states:
- Ready to fix. A repro, a severity, and an owner for this cycle or the next.
- Needs a better report. Back to the reporter with one question. Point them at your bug report template so the missing fields are obvious.
- Duplicate. Linked to the canonical ticket, and the newer copy is closed.
- Not a bug. Moved to the request queue, or closed with the product reason.
- Will not fix. Closed, with the reason on the ticket.
"We'll look at it" does not count. If a bug has the same owner and the same stage afterward, nobody reviewed it.
Bug prioritization without a scoring ritual
Written bug prioritization uses two axes, applied the same way every week.
Severity is the impact if the bug is real: data loss, a security exposure, a broken core path, a slow workaround, or something cosmetic. One noisy customer does not raise severity. An important account can raise priority, which lives in a separate field.
Priority is when you do the work. It weighs severity, how many people hit the bug, whether a workaround exists, and what is already committed. Four levels are enough. Extra levels look precise and then get ignored.
| Level | Use it when | Typical response |
|---|---|---|
| P1 | Production is down, data is wrong, or a security hole is open. | Fix now. Confirm the write-up next meeting. |
| P2 | A core path is broken and there is no reasonable workaround. | This cycle. Name an owner before you move on. |
| P3 | Real pain, a workaround exists, or few people hit it. | Next cycle, unless the stale pass says otherwise. |
| P4 | Cosmetic, an edge case, or a bug you accept. | Backlog. Close at 30 days if nobody still wants it. |
Store the level in a field. A [P2] title prefix goes stale as soon as someone changes their mind.
- Three P1s in a normal week means the scale is inflated.
- Customer tier raises priority one level and leaves severity alone. A blurry logo is still cosmetic.
- A regression from the last release outranks an older bug of the same severity.
- "Quick fix" is not a priority. Small bugs can fill a whole cycle without moving any user goal, so batch them.
Story points can wait until someone picks the ticket up.
A worked example
Northwind is six people on a billing app, with fourteen bugs in the queue. Four of them show the pattern.
Wrong tax on EU invoices. Screenshot, account id, time, and staging confirms it. Wrong numbers on a core path are P1 until someone checks live invoices, so on-call owns it before the meeting moves on. If only drafts are affected, it drops to P2 the same day with a note why.
CSV export does nothing in Safari. Clear steps, Safari 17 only. P3, next cycle, whoever last touched exports.
Blurry logo on the PDF. One customer, amounts correct. P4, no design work in the meeting. If nobody touches it, the stale pass closes it.
"Add a memo field." A request that arrived on a form labeled "report a problem." It moves to the request queue with one line of reason. On the bug list it would only inflate the count.
At minute 32, eleven bugs have had no comment in 30 days. Seven close as will-not-fix: old browsers, a retired plan, duplicates of shipped work. Two get a cycle and two go back for a repro. The list ends shorter than it started.
How the backlog stays honest between meetings
One meeting cannot clean a queue that works like a junk drawer the other six days.
Intake
A report needs four facts before it reaches "needs triage": what the user did, what they expected, what happened, and where (environment, account, time). A short bug report is enough. "Billing is broken" goes back.
Chat messages and alerts can become tickets, but a person still has to accept them. A stack trace tells you where to look. It doesn't tell you the priority. If your bug tracking software opens cards from chat or an error tracker, make "unreviewed" the first stage so those cards can't enter a cycle by default.
One canonical ticket
Link repeats to the original and close the copy. Customer context goes on the ticket you will actually fix. With two open copies, the work happens twice or not at all.
Stages that match the work
Unreviewed, needs info, ready, in progress, in review, done, won't fix. Add a stage only when the team already works that way. Stages like "discussing" and "revisiting" hide tickets that have no owner.
Move the stage when the work changes. Moving a bug to needs info, or to won't fix, is a decision. Adding a comment and keeping the same owner is how a bug sits for a quarter.
The 30-day rule
No comment, no repro, and no schedule for 30 days means the bug is in nobody's plan. Close it or schedule it. Support can answer from a closed ticket that states the reason; a stuck open ticket only delays that answer. If it returns, reopen it. A fresh report beats one left "open" since last year.
Rollover
Unfinished work returns through this review instead of sliding into the next cycle alone. Recommit it with an owner, or drop it. Otherwise every P3 stays open forever.
Numbers worth reading
- Bugs still unreviewed after seven days: whether the meeting works.
- Age of the oldest P2: what support feels.
- Won't-fix versus fixed. A team that never closes anything as won't-fix is hoarding.
- P2 time from "ready" to done, as a percentile, so one stuck ticket can't hide in an average.
If tickets get stamped "ready" at creation, cycle time becomes fiction. Forecast later, from real dates.
Where the work should live
You need an unreviewed stage, a severity field, an owner, and duplicate links. A spreadsheet breaks once two people edit it, and chat loses the decision.
Jira bug tracking can do this, and fails with a dozen statuses, a priority field nobody believes, and a filter so wide there is no queue. GitHub Issues fits a small team in the repo, then strains once support mail and error reports share one queue. Pick from what an issue tracker must record. A new product won't run the meeting for you.
Falrow keeps review as its own stage, turns Slack messages and Sentry errors into tickets, and carries unfinished work forward only after someone has looked. The decisions match on Falrow, Jira, or GitHub. If you're staying with your current tool, the fields for tickets, triage, and workflow rules are still worth a look.
What good looks like after a month
- Unreviewed is empty, or you can name the overflow and the day you'll get to it.
- Every open P1 and P2 has an owner and a cycle.
- Support can answer "when?" from the ticket.
- The board shrank in at least half the weeks, because duplicates and won't-fix items closed.
- A new engineer can see why a ticket is P3 without watching a recording.
If the queue only grows, you're reading bugs without triaging them. Send thin reports back and close what you won't do.
FAQ
How often should we run bug triage?
Once a week for most product teams, plus a short pass the day after a release while regressions are still arriving. Daily triage turns into whoever is online glancing at Slack, and severity follows the loudest thread. P1s don't wait for the calendar: name an owner now and confirm the written reason at the next meeting.
Who should own the triage process?
One person owns the calendar, the queue, and the notes. Rotate the advising engineer, not the lead, or the rules drift. The lead doesn't need to be senior, but they do need to close tickets and cut off debugging. Product and support advise. They don't keep a second priority list in a spreadsheet.
How is this different from sprint planning?
This meeting decides whether a report is a real bug and how bad it is. Sprint planning decides what fits alongside features and chores. The Scrum Guide treats sprint planning as the plan for the Sprint, not a screen for every defect. A bug that skipped the weekly review shouldn't enter that plan. Merge the two meetings and both run long, and the reason a bug was turned down gets lost.
Should we review tiny bugs too?
Yes, for a few seconds each. Check for a duplicate, then decide yes or no. A tiny bug you skip still sits in the count you show leadership. "P4, no owner, look again at the stale pass" is a complete decision. Small fixes you do plan go in a batch, on a cycle, on purpose. Leaving them open "just in case" costs more than closing them.