An architecture improvement backlog has 14 items and capacity for three this quarter. Why does scoring the items fail to prevent the same argument next quarter, and what should sit beside each item instead?

Why a score does not survive the quarter

A score is a snapshot of opinion, compressed into a number that hides its inputs. Three months later the inputs have changed and nobody can tell whether the ranking moved because the world moved or because a more persistent advocate arrived. So the list is re-argued from scratch, by the same people, with the same information, and the quarterly planning meeting becomes a rhetoric contest in which the loudest sponsor wins and the item with no sponsor never surfaces.

The mechanism is simple: a score is not falsifiable. "Value 8, effort 5" cannot be checked against anything. Two people who disagree about a score have no way to resolve it except by seniority.

What should sit beside each item

A trigger: the observation that would make this item the most important one on the list. Three parts, one line.

Replace the shared job runner. Trigger: queue wait at p95 above 90 seconds for two consecutive weeks, or a second team blocked on it in a quarter. Signal: the jobs-queue dashboard. Owner: platform lead.

Now the item does not need an advocate. It raises its hand when the condition is met, and if the condition is never met, it was correctly deprioritised. The argument moves from "how important is this" to "has this happened", which is a question with an answer. It also tells advocates what to do instead of lobbying: go and measure.

What this changes in practice

Three things, every time this is done on a real list.

• Items with no statable trigger get deleted, not deferred. Typically two to four of fourteen. An item whose urgency cannot be tied to any observation is a preference, and keeping it on the list costs attention every quarter forever.
• Two or three items turn out to have already fired, and nobody noticed, because nothing was watching. That is usually the most valuable output of the exercise.
• The list stops needing a full re-rank. Next quarter you check triggers, which takes minutes, instead of re-running the scoring workshop.

Writing triggers for 14 items takes about 90 minutes, mostly spent discovering that four of them have no measurable signal at all — which is itself the finding.

When a score is enough

Sequencing five items inside a single quarter, with one decision maker and no expectation that the list outlives the quarter: just rank them and start. Triggers earn their cost when a list persists across planning cycles, which is the normal condition for architectural work, because most of these items will sit unfunded for a year or more. The failure a trigger prevents is the slow one — an item that is genuinely urgent in month seven being argued about on its original merits in month twelve.

Common weak answers

• "Use a weighted scoring model with more criteria." More criteria make the number harder to challenge, not more accurate. The problem is not the formula's resolution, it is that the output cannot be checked against the world.
• "Let the business decide the priority." The business can rank outcomes it can see. Most architectural work shows up as absence — the incident that did not happen, the migration that stayed cheap — so handing the whole list over guarantees that the invisible items lose. The trigger is what makes an invisible item visible at the moment it matters.