The Not-To-Monitor List: Deciding What Deserves Silence
Ask an admin to show you their monitoring and they will show you what they watch: the dashboards, the checks, the alert rules. Ask them what they deliberately do not watch and you get a pause, then a shrug. The unmonitored things are whatever fell through the cracks. I gave that shrug for years, and I gave it with some confidence. The most underrated document in monitoring is a short, boring list of the things you have chosen not to watch. This is silence selected on purpose, written down, signed, and defensible.
That sounds like paperwork for its own sake. It is the opposite. It is the difference between engineering and luck. The difference only shows on the day the unmonitored thing breaks, which is the one day you cannot write it then. By the end of this guide you will have the one test that sorts monitoring candidates from noise candidates. You will know the categories of systems that usually earn silence (and the exceptions that flip each one). You will have a template for the list itself with filled-in examples. You will also have a way to start yours today with five honest entries.
In this guide
- Why explicit silence beats implicit silence
- The actionability test
- Categories that usually earn silence
- A walkthrough: the mystery device decision
- The parking lot for maybes
- The list's anatomy, and where it lives
- The failure mode: the dumping ground
- Start yours today
- Where to go next
Why explicit silence beats implicit silence
Every network has unmonitored things — that is not a failure, it is scope. The question is whether each gap exists on purpose. There are exactly two kinds of unmonitored system: the one you decided not to watch, and the one you forgot to watch. From the outside they look identical — both are silent. But one is engineering and the other is negligence. You only find out which is which when the thing breaks.
Play that discovery forward. An unmonitored box fails, causes a mess, and someone senior asks the inevitable question: "why weren't we watching this?" With no list, the honest answer is "it never came up," which sounds — and is — careless. The aftermath becomes a blame hunt that usually ends with the worst possible policy: monitor absolutely everything from now on, a cure with its own disease (see data hoarding). With a list, the answer is: "we assessed it, here is the reasoning, here is who signed off, here is the revisit date." The conversation shifts instantly from who failed? to has the situation changed since we decided? That is a review, not a trial. The list is cheap insurance for your team's credibility, and it converts hindsight from a weapon into an input.
Harbourview Clinic ran a discovery scan. This is the sweep a monitoring tool makes to find every address that answers. It came back offering to monitor four thousand one hundred things. The admin did the sensible thing, ticked the servers he recognised, left the unknowns for later, and never got back to later. One of the unknowns was the appointment-reminder gateway, whose hostname was its own serial number. It failed on a Tuesday, and the clinic found out from the patients who did not arrive. When the practice manager asked why nobody was watching it, the honest answer was "it never came up". The review treated that answer the way reviews do. One row in a parking lot would have turned the same meeting into a review of a decision instead of a search for a culprit. The row could have read: "unknown devices from the scan, revisit in thirty days".
The same logic protects you in audits and insurance questionnaires. There, "show us your monitoring coverage" is answered by a scope decision with names and dates. That is far more convincing than a nervous tour of whatever happens to be configured.
There is a quieter benefit too: writing the reason down forces you to actually have one. Half the value of the not-to-monitor list is that some candidates fail the writing test. You start typing "we don't monitor the old NAS because..." and realize you cannot finish the sentence. That means the NAS just talked its way into monitoring.
The actionability test
One question does most of the sorting, and it is worth engraving somewhere visible: if this alert fired at 9 a.m. tomorrow, would anyone do anything? Not "could," not "should in an ideal world" — would, given your actual team, workload, and priorities. The 9 a.m. framing is deliberate: it strips away the 3 a.m. drama and asks the question at your most staffed, most caffeinated hour. If the answer is "no" even then, no alert schedule will ever make this check useful. An alert that produces no action is not monitoring; it is a subscription to interruptions. Walk some familiar candidates through the test and watch it work:
- The test VM. It goes down at 9 a.m. Would anyone act? No — its owner would restart it whenever they next need it, which is the entire point of a test box. Fails the test; onto the list. (Unless the name is no longer accurate — see the exceptions below.)
- The server scheduled for decommissioning. It is being switched off for good next month. An alert today would change nothing. Fails the test — with a twist: the list entry gets a revisit date after the decommission date, to confirm it actually happened.
- The vendor-managed appliance you are not allowed to touch. Its internals alert; you can do nothing but forward the alert to the vendor, who has their own monitoring. Internal health checks fail the test. But note the trap: reachability of that appliance passes the test easily, because "it is unreachable" is actionable — you call the vendor. This split gets its own section below.
- The redundant node whose sibling is watched. Careful with this one. If node B failing changes nothing today, monitoring B's services may fail the test. But B's existence as spare capacity is exactly what you rely on. A dead spare is only discovered when the primary fails. B's basic health passes the test after all: the action is "restore the redundancy before it is needed." What you can skip is alerting on B with the same urgency as A.
Two of those rows deserve a second look. The test VM fails the test unless its name is a lie, which happens more than anyone admits. The decommissioning entry gets its revisit date because an alert on a server that is being switched off next month produces, at most, a wry smile. And decommission schedules have a way of becoming aspirational. And the spare node is watched because a dead spare discovered during the primary's funeral is the classic redundancy tragedy.
Notice the test never asks whether a system is important, expensive, or interesting — only whether an alert about it would change anyone's behavior. That is what keeps the list honest and the argument short.
Categories that usually earn silence
Certain categories come up on every not-to-monitor list. For each, here is the standard reasoning — and the exception that flips it, because every one of these has a flip.
Personal and test equipment
Laptops, lab gear, sandbox VMs, the machine under the intern's desk. Nobody depends on them, their owners notice problems before any alert would, and their "outages" are usually someone rebooting on purpose. The flip: the test box quietly running something real. The tiering exercise in criticality tiers tells the story of test-app-02, which served production shipping labels for two years. Classify by what runs on it today, not by its name.
Systems under someone else's contract
Examples are the firewall run by your MSP (managed service provider) and the leased multifunction copier with vendor telemetry. The MSP is the outside company you pay to look after the firewall. Another example is the hosted PBX (the phone system). Someone else is paid to watch their internals; duplicating their monitoring means duplicate alerts and arguments about whose number is right. The flip — and it is a big one: monitor reachability even when management is theirs. "Is the firewall passing traffic?" is your business continuity, not their contract line. The clean split: their job is the box's internals. And your job is whether your network can still get through it. Plus, ideally, your job includes a check that their service is meeting the promise you pay for. Outsourcing management does not outsource caring.
Things with no response path
If nobody can act on an alert — no access, no authority, no spare, no vendor to call — then alerting is theater. An alert is a request for action; sending one to people who cannot act is just scheduled frustration. The flip: "no action possible right now" is not "no value ever." Consider a daily or weekly digest instead of alerts: awareness without interruption. The branch office link you cannot fix still shapes the conversation with the ISP at contract renewal — if you have the history.
Duplicate perspectives
Do not watch the same truth five ways. Suppose the hypervisor reports a VM's health, the guest agent reports it, and a ping confirms it. Suppose two dashboards graph it too. In that case, four of those exist only to disagree with the fifth during an incident. Redundant checks also multiply alert volume for a single event — one dead VM should not page you four times. Pick the perspective closest to what users experience, keep one independent cross-check for the truly critical, and put the rest on the list. The flip: genuine end-to-end checks are not duplicates. "The server is up" and "users can actually load the login page" sound similar and are different facts; keep both.
Things below the annoyance floor
Some failures cost less than the monitoring would. A meeting-room display that needs a monthly power cycle, the decorative lobby screen, the guest Wi-Fi's captive portal cosmetics. If the lifetime cost of failures is lower than the lifetime cost of checks, alerts, and false alarms, silence is the correct engineering choice. Write that down as a sentence, because it will be challenged. And "we decided the screen is cheaper broken" is a defensible position only when someone decided it. The flip: annoyances that touch customers or executives have a political cost above their technical one. The lobby screen the CEO walks past every morning may buy its way into a quiet daily check. That is not because the tech matters, but because the questions do.
A walkthrough: the mystery device decision
Here is the whole discipline compressed into one routine event. Your discovery scan turns up a device you do not recognize: 10.0.40.117, answering pings, hostname unhelpful. The undisciplined responses are to ignore it (implicit silence) or to slap a default check on it (a small donation to the hoard). The disciplined response is a five-step decision that takes about ten minutes:
- 1. Identify it. MAC vendor lookup, switch port, asking around. It turns out to be the warehouse's temperature logger for the cold-storage room, installed by the facilities contractor.
- 2. Find the stakeholder. Facilities own it; the contractor maintains it; the warehouse manager cares about what it measures.
- 3. Run the actionability test. If "logger unreachable" fired at 9 a.m., would anyone act? Ask the stakeholder, not yourself. The warehouse manager's answer: emphatically yes — cold-storage temperature excursions spoil stock worth real money. And the logger's own alarm only beeps locally in a room nobody sits in.
- 4. Check the consequence, not the box. What looked like a Tier 3 gadget is attached to a genuinely expensive failure mode. The tiering logic from criticality tiers applies to what it protects, not what it costs.
- 5. Write the decision down, whichever way it goes. Here: monitor reachability and, since the logger speaks SNMP, the temperature itself. Had the answers gone the other way, it would have gone onto the not-to-monitor list. That alternative would have been a contractor-monitored device with its own alerting and a documented response path. Its list entry would have had exactly that reason and the contractor named in the decided-by column.
Ten minutes, one row written somewhere permanent, zero future mystery. Every unknown device deserves this exact treatment, and the not-to-monitor list is where half the outcomes land.
The parking lot for maybes
Some candidates refuse to sort cleanly. One example is the new SaaS tool one team is piloting. This is software rented over the internet rather than installed on your servers. Another is the legacy app that might be retired "soon." Another is the switch in the acquired office you have not audited yet. The wrong moves are to monitor them by default (hello, hoard) or to forget them (hello, negligence). The right move is a third state: the parking lot. This is a short section of the same document for things that are neither monitored nor forgotten. Each has a date on which the question will be asked again. "Undecided until the pilot ends" is a perfectly respectable engineering position; "undecided forever because nobody wrote it down" is not. The date is what separates a parking lot from a memory hole.
Keep the parking lot short and the dates near. If an item has been parked through three revisit dates, that is the document telling you something. Either nobody cares, so it belongs on the not-list with that as the honest reason. Or everyone is avoiding a hard conversation, so schedule the conversation. A parking lot is a waiting room, not a residence.
The list's anatomy, and where it lives
The list itself is deliberately tiny — four columns, one row per decision. Anything more elaborate will not get maintained, and a not-to-monitor list that is not maintained is just a forgot-to-monitor list with formatting.
| Item | Why not monitored | Decided by | Revisit |
|---|---|---|---|
| Lab VMs (vlan 60, all) | Test workloads; owners self-serve; no user impact | IT + eng lead | Yearly |
| SRV-LEGACY-01 internals | Decommission scheduled; data already migrated | IT + ops mgr | 1 month after decom date — confirm it is actually off |
| MSP firewall internals | Under MSP contract; we watch reachability + throughput only (the split) | IT + MSP contact | At contract renewal |
| Node B service checks (cluster FS2) | Redundant sibling; basic health watched at low severity, services alert only on A | IT | Yearly |
| Meeting room displays | Below the annoyance floor; failures self-reported within hours | IT + office mgr | Yearly |
| Parking lot: pilot CRM sandbox | Undecided until pilot verdict | IT | End of pilot (dated) |
Two details in those rows do disproportionate work. The decided by column is the sign-off — the name that turns "IT ignored it" into "we agreed." The column should include a non-IT stakeholder whenever the silence touches their world. The revisit column is what keeps decisions from fossilizing. Every reason on the list was true on the day it was written. The revisit date is the scheduled check on whether it still is. A reason with no revisit date stays true forever, which is more than can be said for the server.
Where does the list live? Next to your scope document. The two are halves of one statement, "here is what we watch and why, here is what we don't and why." Together they are the practical output of the goals conversation in monitoring goals and scope. And the list earns its keep in one specific meeting: the coverage review, where you diff your inventory against what is actually monitored. Every gap that diff surfaces gets reconciled against this list. A gap on the list is verified as still-intentional. A gap not on the list is a real finding that gets monitored or gets an entry. Without the list, every coverage review re-argues every gap from scratch. With it, the review takes half the time and produces decisions instead of déjà vu.
Tip: record not-to-monitor decisions in the monitoring tool as well, wherever it lets you annotate or tag a discovered-but-unmonitored device. For example, put a tag like no-monitor: see scope doc on the device entry. The next admin who stumbles across it in a discovery scan learns in five seconds that the silence is deliberate. That prevents them from "fixing" it and re-creating the noise you pruned on purpose.
The failure mode: the dumping ground
The list has one way to go bad. Once it exists, it becomes tempting to toss things onto it to avoid work. There is the flaky check you never tuned. There is the awkward device that needs SNMP credentials from a vendor who never replies. There is the system whose owner is annoying to schedule with. "It's on the not-to-monitor list" starts to mean "we gave up," and the document that was supposed to certify deliberate silence begins laundering neglect instead.
The defense is a simple standard for entry: every reason must be one that would survive being read aloud after an outage. Imagine the unmonitored thing has just failed badly, and your entry is on the screen in the review meeting. "Test workload, no user impact, owner self-serves" survives. "Vendor-managed; we watch reachability, internals are contractually theirs" survives. "Monitoring it was fiddly" does not survive, and neither does "the check kept false-alarming so we stopped watching." That one is not a scope decision. It is an unfixed check wearing a scope decision's clothes. If the honest reason is "hard" or "annoying," the item does not belong on the list; it belongs on your backlog. The list documents silence you would defend, not silence you would explain away.
Start yours today
This document costs less than any other artifact in monitoring: no software, no budget, no meeting series. Open a page next to your scope doc, draw the four columns, and write five honest entries before the day ends. Most admins can name five deliberately-unwatched things without standing up: the lab VLAN, the doomed server, the vendor's box, the redundant sibling, and the meeting room screens. Add names, add revisit dates, and show it to one stakeholder for the sign-off column. Twenty minutes, and the next time something unmonitored breaks, you will be holding engineering instead of a shrug.
Where to go next
The not-to-monitor list is one corner of a single discipline: deciding where attention goes, and writing the decision down so it outlives the person who made it. Criticality tiers ranks what you do watch. And dependency thinking finds the hidden things that must never land on the not-list. Meanwhile, anti-hoarding prunes what stopped earning its place. And the intake rule makes the watch-or-skip decision part of every deployment so the list stays current without heroics. The rest of Monitoring Scope: What to Watch (and Skip) ties the whole practice together. Five rows, four columns, twenty minutes. The next time someone asks what you deliberately do not watch, you will have something better than the shrug.
Put this into practice with SysPeep Monitor
A deliberate not-to-monitor list starts with knowing everything that exists — you cannot choose silence for a device you never found. SysPeep Monitor auto-discovers the devices and interfaces on your network. So your watch-or-skip decisions are made against a complete picture. SysPeep Monitor's alert rules keep the watched side quiet enough that the silence stays meaningful. The free trial installs on any Windows machine.
Download Free TrialLearn About SysPeep