What makes an alert worth paging for

The disk really will fill before morning. The alert wakes someone who can verify the graph but cannot add space, and the authorized specialist is unavailable. The measurement is right. The missing piece is a person who can do something useful before time runs out.

Review the alert around that response, not merely its contribution to page count. What would waiting cost, who could intervene and is the signal credible enough to justify the interruption? The result might be a page, a staffed queue or an unresolved response gap.

The recovery option that waiting would remove

The awakened responder has two times to compare: when the disk is likely to fill and when someone able to help will next be available. Waiting may remove the chance to add space before requests fail. Capacity exhaustion is one such deadline; deteriorating service or a closing recovery window can create others before an outage begins.

The responder cannot use the whole remaining window for the capacity change. Some of it goes into checking the growth, some into obtaining permission, and some into carrying out the work. If those steps simply need more time, an earlier page may help. If the permission owner cannot be reached at all, moving the same page earlier leaves the responder waiting for the same person.

That is why a correct graph is only part of the decision to wake someone. The projected wait must have a meaningful cost, the recipient must be able to reach a useful intervention, and the signal must be credible enough to warrant the work. Our responder has established that the disk is filling. They have not yet established that this page gives them a way to prevent it.

Google's guidance considers precision, recall, detection time and reset time. Applied here, we would ask how often the capacity warnings are useful, which real problems they miss, how early they arrive and when they clear. Those checks improve the signal, but the awakened responder still needs enough time and someone authorized to act.

Check urgency and the available response

Read diagram description

Ask whether waiting can lose the chance to prevent harm and whether a useful intervention is reachable in that window. Urgent with a usable response: page the reachable responder. Can wait: route to a staffed work queue. Urgent without a usable response: retain an unresolved coverage gap and repair escalation, delegated scope or the explicit risk decision; do not assume the signal can simply be suppressed.

An accurate warning with nobody able to act

In this hypothetical alert, exhaustion precedes the next staffed period. The primary can verify it, but only an absent specialist can add capacity. Suppression would hide real exposure; repeating the same page would keep waking someone unable to finish the response.

Repair the route if possible: make the specialist reachable, delegate a bounded capacity action or use supported load shedding to gain time. If none is available, the accountable owner must address or explicitly accept that exposure. Fewer interruptions would not make the disk stop filling.

Now change the projection to several staffed periods away under stable growth. A tracked task for the capacity owner may leave enough time. The observation still matters, but the intervention can wait. Revisit the route if growth changes.

Why a total score can hide a disqualifying gap

Our disk warning could score well for its accurate graph, good documentation and reliable delivery. None of those points makes the missing specialist answer. The routing record needs to say which situation we actually have: someone can respond in time, the work can wait for a staffed period, or the coverage needs repair. A high total should not conceal that last answer.

Suppose the same disk warning arrives twice while the responder is looking for the specialist. Recognizing the duplicate can remove a repeated notification. Grouping a different symptom with it is another matter: correlation may combine independent failures, not just copies of one message. The original signals and a way to split the group must remain available, so a quieter channel does not hide a second problem. The correlation guide tests that behavior.

Can the disk warning still reach help?

Suppose a bounded trial changes where the warning goes. Replaying a known disk-filling case lets the team ask whether it still reaches a usable response; normal work and missing telemetry need their own cases too. The page count is only one view. Customer evidence, incident reviews and signals outside the suppression rule can reveal harm that the quieter pager no longer shows. Fewer interruptions help only while important failures remain detectable.

A familiar disk case may be easy to replay. A rare catastrophic failure may offer almost no examples. That absence does not make its warning worthless. Before changing its route, the team still needs to say what remains uncertain, what missing the failure would cost and which independent checks could reveal it.

The limited trial can answer the disk question without settling every rare case. Its record should keep those unknowns and the condition that would reverse the routing change. If the growth changes or a needed warning is missed, the decision needs another look, not a better-looking page-count chart.

An alert routing record

Complete the following record for one alert or genuinely duplicate family. Each field addresses a separate condition in the routing decision.

The storage graph did not change when staffing changed. What changed was whether the warning reached someone who could use the remaining time. A useful review repairs that connection and leaves any unresolved gap visible, alongside the disk still filling.