Remote hardware health checks
what we test and how it prevents failure

Hardware does not fail without warning

Computers almost never die suddenly. A failing drive starts with a handful of sectors that do not read on the first attempt. A dead laptop battery got there by shedding capacity a percent at a time over two years. A machine that seizes up because the disk is full filled at a steady rate over several weeks. The failure is instant; the decay in front of it is measurable.

The problem is not the absence of signals, it is that nobody reads them. A person sitting at the computer sees one data point: right now. They cannot see that two weeks ago the drive had 31% free and today it has 19%.

The XIT agent sends a health snapshot every 15 minutes — 96 readings a day, roughly 5,760 per machine across the 60 days of history we retain. That turns "the disk is filling up" into a number: how many percent per day it is losing, and how many days remain before Windows can no longer install its own updates.

Here is every hardware signal we measure, the threshold that raises an alert, and how much warning each one buys you:

How you know a disk is about to fail

Every modern drive keeps an internal report card called SMART. It counts how many sectors it has had to relocate, how many reads failed, how many hours it has been powered, and how much of its rated life it has spent. This data exists inside every computer on earth. Almost nobody looks at it.

XIT looks at it. Once a day the agent pulls the health status of every physical disk together with its reliability counters: wear percentage, power-on hours, temperature, and total read and write errors. The values are stored on the server and carried forward between daily reads, so the picture never goes blank.

The moment that status leaves healthy, we raise a critical alert naming the specific drive. In the machine's health score it is also the most expensive check to fail: 35 points, more than any other. The reasoning is blunt. A failed drive is the one component that takes your data with it. Antivirus that was switched off can be switched back on. Photos that are gone are gone.

Alongside the status we watch reallocated and pending sectors. Once those counters stop reading zero, the drive is coming apart even if it still describes itself as healthy. That is when a technician calls and says: back up today, we replace this week. Continuous monitoring is what makes catching that moment possible.

Free space, and the date it runs out

A full disk is not an inconvenience, it is a fault. When space runs out, Windows stops installing security updates, the page file cannot grow, applications fail on save, and the machine slows to the point of being unusable. The gap between 15% free and 3% free is the gap between a slow computer and a stopped one.

The agent measures the free percentage on every fixed drive, on every report. Below 20% raises a warning. Below 10% the alert becomes critical. The worst drive sets the status, so a choking second drive is never hidden inside an average.

From "filling up" to "eleven days"

This is where the history earns its keep. We fit a slope across the recent readings — how many percentage points of free space the disk loses per day. If that rate exceeds 0.3% per day, we calculate when the drive will reach 3% free. If that date falls inside the next 60 days, we raise an insight carrying the numbers themselves: days remaining, percent lost per day, and how much is free right now. Under seven days, the alert is critical.

A second stage then kicks in. On any machine with less than 20% free or less than 25 GB free, the agent runs a folder scan once a day: the twelve largest folders on that drive, then one level down inside the largest of them, returning full paths with sizes in gigabytes. On healthy machines the scan never runs, so it costs nothing. Instead of "free up some space" you get "there is 84 GB of an old backup folder on drive D". If your machine already feels sluggish, our guide to a slow computer covers the causes that are not hardware.

Memory: when 85% is fine and when it is a leak

High memory use is not automatically a problem. A machine at 85% while someone works with twenty browser tabs open is behaving as designed. The same number at three in the morning, with nobody touching the keyboard, means something else entirely.

So we run two separate checks. The first is level: above 80% raises a warning, above 90% raises a critical alert. Past 90% the machine leans on the page file on disk, which is orders of magnitude slower than real memory, and that is where the user feels everything stall between applications.

The second is trend. If memory consumption climbs at more than 1.5% per day while the machine is already above 60%, we raise a suspected memory leak. Leaking software never releases memory it has finished with, so the baseline creeps upward: 62% today, 64% tomorrow, 90% in a fortnight. Nobody notices that line, because any single day looks almost identical to the one before.

The distinction matters because the fixes differ. Genuine high load is solved by closing applications or fitting more memory. A leak is solved by identifying the offending program — adding memory only postpones the same wall by a few days.

CPU and heat: a spike is not a problem, sustained load is

Every computer hits 100% CPU several times a day: launching an application, running a scan, loading a heavy web page. Alerting on each of those is noise, and noise is what teaches people to ignore alerts entirely.

So load is measured as an average. We take the recent samples and only raise a sustained-CPU insight if their average exceeds 70%. An average like that does not come from opening a program — it means something is running continuously in the background: a stuck process, a scan that never finishes, mining software, or a service caught in a loop. The notification names the responsible process, because "CPU is busy" on its own tells nobody what to do.

Temperature is measured in parallel. The agent reads CPU temperature and flags the machine as throttling above 90°C. A second measurement catches the same fault even when the sensor reports a low number: the ratio between the clock speed the processor is running at and the speed it is rated for. A CPU below 70% of its rated clock is throttling itself, and the cause is nearly always dust blocking a fan or dried-out thermal paste.

This is a frustrating fault, because nothing crashes and no error appears — the machine is simply slower, gradually, over months. Reading temperature alongside the processor and memory inventory, including what sits in each memory slot at what size and speed, is what lets us say whether a computer is slow because it is hot or because it is underpowered.

Battery: when replacement is actually due

A laptop battery starts losing capacity the day it is bought. The only question is how much. The agent calculates wear from the battery's own figures: the maximum charge it can hold today against the design capacity it was manufactured with. A battery built for 50 units that now holds 30 is 40% worn.

Below 20% wear the state is normal, between 20% and 40% it is fair, and at 40% or above we raise an alert recommending replacement. We also read the charge-cycle count from the system's battery report — the number that explains why a two-year-old laptop that gets unplugged constantly is more worn than a three-year-old machine that mostly sits on mains power.

The value of this number is planning. A battery at 42% wear will not strand you tomorrow, but it will strand you in some meeting over the coming months, and replacing it in a convenient week beats replacing it the day it dies mid-presentation. The on-demand battery check also returns the current charge level, battery status and a full battery report; on Macs, power and sleep configuration too.

A driver error means a real component is dead

A device driver is the software that lets hardware talk to the operating system. When one is in an error state, that is not an abstract technicality. It means an actual component is not working: the printer that will not print, the network card that vanished, the camera that will not open in a call, the audio that stopped.

The agent collects every device reporting an error state, up to fifteen of them, with the name and class of each. If even one device is on that list we raise a warning, and the notification names the device itself rather than using a generic phrase.

Many of these faults surface here before the user has tried to use the component. If a printer driver breaks on Monday after an update, someone who prints once a fortnight discovers it on the day they urgently need to print — and we knew that Monday. The driver check also returns the full list of installed drivers.

The checks that need a restart, and why we never force one

Some hardware tests cannot run while the machine is in use. A real memory test requires the RAM to be empty of the operating system, so it runs outside Windows after a dedicated restart. A deep disk scan locks the volume. These are the highest-weighted tests in the health score — a memory test is worth 30 points and the deep disk checks 25 points each — because they are the only ones that can confirm or rule out a physical fault.

Our position is simple: we recommend restarts, we never force them. Remotely we can read the result of the last memory test that ran on the machine, with its date. Running a fresh one needs a dedicated restart, and that is something we ask for and explain — you choose when. No machine reboots mid-task because of a decision we made.

The file-system integrity scan can run without interrupting anyone: it checks the structure of the file system and returns its summary lines. The deep disk verification examines the volume structure itself. Pairing a physical check (SMART) with a logical one separates a dying drive from repairable file damage — the distinction that decides whether the answer is a new component or a software fix. The same logic drives our crash analysis, which hunts for the driver behind a blue screen.

What happens once something is found

Underneath the alerts sits a quieter layer: the hardware inventory. Which physical disks are installed, their model, size and status; the exact processor; what occupies each memory slot, at what capacity and what speed; the drive type and its protocol; the power and sleep configuration. None of this raises alerts, but without it every recommendation is a guess. You cannot say "add more memory" without knowing whether a slot is free.

When something is found, it does not stop at an alert. Alerts reach a human technician, and findings are summarised in a plain-language report split into six sections, one of which is hardware. Every line is written in terms of what it means for you rather than the name of the tool that produced it. You can see exactly what a full report looks like.

It is worth stating what is not collected. The agent reads system and configuration state. It does not read, collect or upload personal files, documents, photos, email or browsing content. A hardware check does not require access to your content, so it is not given any — the full detail is here.

Pricing is $5 per month per computer for monitoring, or $10 per month per computer for monitoring plus a human technician who acts on what is found. A full breakdown of what each package includes is on the pricing page.

FAQ

Read next

• Continuous monitoring every 15 minutes — the system behind these checks
• Crash and blue-screen analysis — when hardware has already brought the machine down
• Why a computer gets slow — the causes that are not hardware
• Pricing and what is included — $5 and $10 per computer per month