Knight-Ware Labs All Articles
Software Architecture

Fair Flow, Not Fast Closure: Designing Backlogs That Expose Waiting

By Knight-Ware Labs Software Architecture
Fair Flow, Not Fast Closure: Designing Backlogs That Expose Waiting

Photo: Rathfelder, CC BY-SA 4.0, via Wikimedia Commons

Waiting lists are seductive because they aggregate. A total of 7.6 million people waiting sounds like one problem. In practice it is many queues with different clinical urgency, different harm profiles, and different rates of deterioration. Software backlogs have the same defect. A single count of open tickets tells leaders almost nothing about which work is ageing, which work is harmful, and which work is merely old.

For engineering leaders, the lesson is not to copy public-sector bureaucracy. It is to design backlogs as queues with visible service classes. Optimise for flow and fairness, not raw closure.

Aggregates Hide Harm

A backlog can look stable while harm accumulates. If a team closes ten low-value tickets and leaves one security fix untouched, the count improves and the risk worsens. If a platform team clears quick requests while an accessibility defect waits for months, the average looks healthy and the experience for affected users does not.

The first architectural move is to stop treating the backlog as a single list. Separate work by consequence: safety and security, customer-facing defects, regulatory obligations, developer productivity, and discretionary improvement. These categories have different waiting costs. A four-week wait for a cosmetic change is not equivalent to a four-week wait for a data-loss fix.

Triage Is an Architecture Decision

Triage is often described as a process. It is also a design decision about where work enters the system, who can reprioritise it, and what evidence is required. Without triage, the backlog becomes a first-in, first-out queue with political exceptions. With triage, it becomes a set of service classes with explicit rules.

Define entry criteria for each class. State who can assign urgency. Require a harm statement for expedited work. Record the decision. This is not bureaucracy for its own sake; it is the minimum needed to make trade-offs visible.

Service Classes and Category SLAs

Not every ticket needs the same promise. A category-specific service level agreement is more honest than a single target. For example, security vulnerabilities with a known exploit might have a 48-hour triage and a seven-day remediation target. Customer-facing payment defects might have a four-hour response and a two-day fix target. Low-risk enhancements might have a thirty-day review target.

The SLA should describe the wait, not just the work. It should say when the item will be assessed, when it will be updated, and when it will be resolved or re-planned. That distinction matters because many backlogs fail at the waiting stage, not the working stage.

Make Ageing Visible

Queues improve when ageing is visible. Add age bands to dashboards: under seven days, eight to thirty days, thirty-one to ninety days, and over ninety days. Show the oldest item in each service class. Show the median and the tail, not only the average.

A simple board can do this. A more mature system can calculate waiting time from entry to first assessment, from assessment to decision, and from decision to delivery. These are different queues. Collapsing them hides where the delay actually sits.

A Public-Sector-Inspired Scorecard

Engineering leaders can borrow the discipline of public waiting-list reporting without borrowing its scale. A useful scorecard has five parts.

First, demand: how many items entered each service class this week, and from where. Second, flow: how many were closed, and at what stage. Third, fairness: the proportion of high-harm items assessed within their target. Fourth, ageing: the oldest item and the 90th percentile wait by class. Fifth, rework: how many items returned because the fix was incomplete.

Review the scorecard monthly. Ask why the tail is long, not only why the average is acceptable. Ask which classes are being starved by urgent work. Ask whether the urgent work is genuinely urgent or merely loud.

Flow With Fairness

The goal is not to close more tickets. It is to move work through the system in a way that matches consequence and keeps waiting visible. That may mean limiting work in progress, protecting capacity for high-harm classes, and accepting that some low-value requests will wait longer.

A backlog is a queue architecture. Treat it as one. Name the service classes, publish the waiting rules, measure the tail, and review the harm. Throughput matters, but triage comes first. Without it, a team can be busy, fast, and unfair at the same time.