1. LandOffer
  2. Blog
  3. All articles
  4. Bug Severity vs. Priority: Run Triage with Impact and Reproduction Evidence

Engineering Practice

Bug Severity vs. Priority: Run Triage with Impact and Reproduction Evidence

Use six complete fictional defect decisions to separate impact, reproduction confidence, mitigation, and repair order. A real isolated Python fixture demonstrates only its own data-loss branch, while a high-severity export case changes priority as containment evidence changes.

9 min readLandOffer team

Bug severity describes the impact when a defect occurs; priority records what the team will investigate or fix next. Assess them separately using affected tasks, possible loss, exposure, reproducibility, and the quality of any workaround. A serious defect can have a later repair slot when its exposure is genuinely contained, but that decision needs an accountable owner and a review trigger. A low-impact defect does not become severe because someone wants it fixed tomorrow.

This LandOffer.ai guide contains a six-defect triage queue and a completed scheduling revision. The note application, releases, people, reports, and management decisions are fictional. A small Python fixture was actually run offline to illustrate one data-loss mechanism; it is not application code or a product test. Sources were checked October 27, 2026. No customer data, production service, tracker, or release setting was accessed or changed.

Two restorers inspect a cracked vase beside a separate rail of illustrated work tags.

Define the two fields before arguing about a label

Mozilla's defect-severity documentation describes severity as the scope of a bug's effect on Firefox, displayed alongside priority. Its priority definitions separately describe release-cycle scheduling: for example, P1 is the current cycle, while P2 concerns following cycles. These are Mozilla's conventions, not a universal interpretation of the labels in your team's tool.

Agree on local definitions and who may change them. A QA lead may supply an initial impact assessment while a product or release owner accepts scheduling trade-offs. Do not assume that a reporter's urgency label establishes either field or that an engineer's estimate alone determines acceptable exposure.

The fictional team below uses High for potential data loss or a blocked essential task, Medium for a substantial limitation with context still needed, and Low for a cosmetic defect without functional loss. Unknown remains available while impact is investigated. These are teaching definitions, not mappings to Mozilla's S1–S4 scale.

Its separate scheduling terms are Now, Next slot, and Investigate by a named review. None means a fix has shipped. Every decision includes an owner, an evidence gap, and a condition that could change the order. Treat a calendar target as a request until the named owner accepts the work and capacity is available.

Establish impact and confidence without mixing them

Describe the affected task and the consequence if it fails. Losing note content differs from displaying an old product name on an export cover. Then establish what is known about exposure: released or prerelease, default or optional path, supported environments, affected configurations, and whether the claimed workaround is usable by the people doing the task.

Reproduction evidence controls confidence and the next investigation. It is not a customer-incidence estimate. Three successful reproductions in a deliberately selected configuration do not mean three percent or all users are affected. A single credible report can still warrant prompt investigation when the potential consequence is serious.

Mozilla's bug-writing guidance asks for precise steps, expected and actual results, environment information, and separation of observation from speculation. Use those principles to evaluate existing reports during triage. The question here is which evidence changes impact or ordering across the queue, rather than how to write every field of a new bug report.

Keep unknowns explicit. “Could not reproduce on this build with these inputs” is narrower than “does not happen.” Likewise, a note about customer urgency is evidence of a deadline or concern, not proof that the underlying technical failure is widespread. If an issue may require a separate security or incident process, route it through the authorized process instead of waiting for ordinary backlog discussion.

Six completed triage decisions for one fictional release

The team is reviewing an invented desktop note-outline application, build N4, on November 2. All application behavior and reproduction reports in this table are authored scenario inputs. Only the isolated fixture in the following section was actually executed. Release owner Yusuf, engineering lead Neri, and QA lead Imani agree on the assignments inside the story.

Defect and impact evidence in the scenario Severity and confidence Scheduling decision, accepted owner, and next gate
B1: saving an untitled note can replace its body with empty text. No reliable recovery is established High; reported destructive path, with a similar mechanism demonstrated only in the teaching fixture Now: Neri takes the candidate save-path investigation today. Yusuf holds the affected save-path release until application regression evidence is reviewed
B2: a prerelease export format omits endnotes. Three internal testers used that format; it is not enabled in the released build High when triggered; narrow known exposure, broader scope not established Next slot: Asha accepts investigation starting November 4. Yusuf keeps that format out of release and reviews containment November 3; no permission to enable it while unresolved
B3: keyboard focus becomes trapped in the print preview in a supported configuration; no adequate keyboard escape is documented High; blocked essential task in the supplied scenario Now: Imani owns confirming the exact supported configuration and routing a fix with the UI engineer today. A mouse-only exit is not accepted as an adequate keyboard workaround
B4: the export cover displays an old product name, with note content intact; a demonstration is tomorrow Low; cosmetic effect, deadline confirmed by product owner Next slot in parallel: Jonah accepts the cover-label correction and review. His available documentation/UI capacity does not replace B1/B3 investigation time
B5: drag reordering sometimes moves an outline item to an unexpected position; the original order can be restored in the reported case Medium, provisional; intermittent steps and scope incomplete Investigate by November 4: Mei collects the minimal input/state and checks persistence. Evidence of unrecoverable loss would reopen the impact assessment immediately
B6: a report says reconnecting sync produced duplicate outline entries; whether any original entry was overwritten is unknown Unknown; possible duplicate symptom, destructive impact neither confirmed nor excluded Investigate today: Ravi accepts a bounded evidence review. Yusuf will reassess at the next daily check; lack of reproduction does not send it silently to backlog

The queue is a decision record, not a score that automatically sorts tickets. B4 can receive a small parallel slot before B2 because the latter's risky export remains contained and its investigation has an accepted owner. B4's demonstration deadline has not changed its severity. Revisit the plan if that parallel work diverts the people needed for uncontained data loss or keyboard access.

The export containment also has a cost: the new format cannot be offered. Record that limitation rather than describing a disabled feature as a complete user workaround. Yusuf owns that release restriction; Asha does not have unilateral authority to accept the product risk by starting a repair later.

A hand holds a glass cover over one cracked bowl while another slides an ochre tag on a curved rail beside an uncovered cracked bowl.

Use a reduced reproduction without overstating what it proves

For B1, the article's actual local check used an intentionally defective pure-memory Python function. It returns a new dictionary and writes no note file. The essential case is:

def legacy_save(note):
    return {"title": note["title"],
            "body": note["body"] if note["title"].strip() else ""}

note = {"title": "", "body": "alpha\nbeta"}
print(repr(legacy_save(note)["body"]))

The observed output was ''; the expected body was 'alpha\nbeta'. The empty-title case returned an empty body in three repeated calls. A whitespace-only title also lost the authored body, while a named note retained it. These observations were recorded with Python 3.12.14 on macOS; they concern this small function and these inputs, not persistence, sync, operating-system coverage, or any real note product.

A candidate function that returns the supplied body unchanged preserved all three tested bodies: empty-title, whitespace-title, and named-note cases. That is bounded evidence for the candidate behavior, not proof of an application fix. It does not establish recovery of already lost notes or the correct title-validation policy. The release hold remains until the real application's authorized checks cover its relevant save and recovery paths.

This is why a triage record should distinguish a report, a reduced mechanism, an application reproduction, and a verified repair. The actual fixture demonstrates body loss in its authored branch; confidence about an application requires evidence from that application. It cannot establish how many users reach that branch or whether the fictional application's implementation shares it. A plausible code explanation should not quietly become a confirmed root cause in a different system.

For intermittent B5 and uncertain B6, choose a bounded next check instead of inventing reproduction success. Preserve the reported environment, state, and observation separately from a suspected cause. A request for more evidence should name an owner and review point, so an unresolved loss risk does not become an unmonitored waiting state.

Keep high severity while changing the repair order

At the November 3 review in the fictional case, Yusuf confirms that B2's export format is still excluded from the released build and that no activation is approved. Asha has accepted the November 4 investigation slot. B1 and B3 investigations remain active. B4's small correction is reviewed, but its publication is not recorded as complete. The team retains B2 at High rather than lowering the label to make the queue look consistent with its later slot.

The completed decision reads:

B2 can omit content when the prerelease format is used, so its severity remains High. Its repair investigation starts after the current B1/B3 assignments because the affected export is not released and is held from activation. Yusuf owns confirming that containment; Asha owns the next investigation. Review again November 4 before considering any enablement. If the format is found in a released build, if the same omission affects another format, or if the restriction cannot be maintained, reopen the priority immediately. No release approval follows from this schedule.

At the next authored check, QA finds an omission in a second export path whose release status is not yet verified. The team does not claim containment has definitely failed, or that all exports are affected. Yusuf pauses the assumption supporting the later slot and asks Imani to verify release exposure. Asha’s investigation moves to Now. Yusuf accepts deferring her routine parser-refactoring review to make room, while the existing B1/B3 owners retain their urgent assignments.

The new finding changes priority before its full scope is known because it undermines the reason for deferral. Severity remains High. The record now says “exposure under investigation,” retains both paths separately, and requires an updated decision after the verification. No fix, recovery, or safe release is declared.

Mozilla's triage policy names triage ownership, calls for explaining severity changes, and describes reassessment of important bugs and requests for missing information. Its specific assignment windows and release rules are Firefox policy, not deadlines for this case. The transferable point is an accountable, revisitable decision with a recorded rationale.

End triage with decisions people can act on

Close each item with distinct statements: impact assessment and confidence; ordering decision and accepted owner; evidence still needed; mitigation authority; and a review trigger. If there is not enough information to classify impact, record that state while assigning the investigation. Do not convert missing data into low severity or an unowned promise to look later.

Explain the trade-off using exposure, task impact, and accepted capacity rather than the force of a request. An intermittent defect can deserve early investigation, a high-impact contained defect can await a bounded slot, and a small fix can proceed in parallel without changing either impact judgment. Those decisions remain conditional on their evidence and the actual resources available.

Use the six-row record to review accepted defects, confirm who owns each next check, and name the fact or review date that would reopen its ordering. LandOffer.ai publishes this engineering guide; it does not run your triage or verify your release. A completed triage discussion produces accountable next actions and visible uncertainty, not a promise that the product is safe to ship.

Sources and Further Reading