You are responsible for reliability standards across 60 teams holding roughly 4,000 alert rules, and you have no authority to edit another team's rules. How would you drive pager noise down across the whole organisation?

## Why the per-rule process does not scale here A single team can sit down with 90 days of firing data and work a ranked list. Sixty teams cannot be worked that way by one central group: 4,000 rules is more context than any platform team can hold, and the person who knows whether a rule is worth keeping is on the team that wrote it. The central role is not to make the decisions — it is to change the conditions under which sixty teams make them. The honest constraint in the question is the important one: **no authority to edit another team's rules.** Any answer that assumes you can mass-delete is answering a different question. ## Lever 1: measurement and visibility You can almost always instrument what you cannot control. Central paging infrastructure sees every notification, so you can publish, per team: - **pages per on-call shift**, especially out-of-hours pages; - **actionability rate**, if acknowledgement captures it; - **the noise concentration** — how much of a team's volume comes from its top three rules, which is usually most of it and makes the fix look tractable; - **the detection ratio** — incidents found by an alert versus reported by a customer. Publishing these side by side does most of the work, because a team seeing itself at 60 pages a shift next to peers at 3 does not need to be told anything further. Present it as a health signal, not a league table to punish with, or teams will optimise the number instead of the pager. ## Lever 2: defaults through the platform The highest-leverage central work is making the good configuration the one teams get for free: - **shared routing with sane grouping** so nobody has to discover grouping keys themselves, and a per-instance fan-out cannot reach a pager as forty notifications; - **dependency suppression for shared infrastructure** — the datastore, cluster control plane and network alerts you own centrally can inhibit downstream symptoms across all sixty teams from one rule set; - **silences that expire by policy**, with a maximum duration and automatic reporting of long-lived ones, so blind spots cannot accumulate quietly; - **rule templates** that carry a pending duration, an owner label and a runbook annotation by default; - **a lint in the delivery pipeline** — a paging-severity rule with no owner and no runbook link fails the check. This is enforcement without authority over content: you are not saying what may be alerted on, only that a page must name who and what. Each of these reduces noise for teams that never asked, which is the only kind of central intervention that scales. ## Lever 3: a budget, not an audit A reliability organisation already knows this shape from budgeted objectives: set the target, let the owning team choose how to meet it, and attach a commitment to what happens when it is missed. Applied to the pager, a team owns a load target — say, no more than a small handful of pages per 12-hour shift, and out-of-hours pages counted more heavily — and sustained breach commits them to spend the following cycle on alert quality rather than roadmap work. The commitment is what makes it real. A target with no consequence is a dashboard. The consequence does not need to be punitive; it needs to be pre-agreed with the team's own leadership, so that when it triggers it is a policy taking effect rather than an argument starting. ## The trap, and the counter-metric **A pager-load budget on its own rewards under-alerting.** The fastest way to hit any page target is to delete the rules, and a team under delivery pressure will find that path. This is the single most important thing to say about org-wide noise governance. The pairing that prevents it is the detection ratio: what share of a team's incidents were detected by an alert rather than reported by a customer or discovered by another team. A team is healthy only when both move in the right direction — pages down and detection holding. One number alone is gameable in an obvious direction, and both together are not. Additional guards: require that pruning decisions are recorded with a reason, and ask postmortems the direct question of whether a removed or suppressed alert would have caught the incident earlier. ## What you accept Some teams will remain noisy. Central deletion of their rules is a fight that costs more political capital than it returns and that ends with a team who no longer talks to you about reliability at all. The workable posture is: make the good path effortless, make the numbers visible, make the budget a pre-agreed commitment, help directly with the two or three worst offenders where you have a relationship — and let the residue be visible rather than hidden. Visible noise in three teams is a manageable state; hidden blindness across sixty is not.