A parts supplier lives or dies on paperwork that matches reality. A fastener without a trail is a liability, and an inspector reviewing records does not ask whether you feel confident; they ask to see the evidence. The same mindset works for the servers behind an online parts catalogue, a stock system or an order desk. That mindset is the core of SysOps, the routine care of server hosts, and what follows converts it into a worksheet that one person can complete in a single afternoon.
The framing here follows the Ops Atlas, whose SysOps entry is explicit that the discipline is measured, not just practised. It lists indicators such as baseline conformance, monitoring coverage and scheduled-job success, which are exactly the sort of figures an auditor can ask for. Use the questions below as a worksheet: for each one, write down the evidence, not the opinion.
How to score the audit
Give each question one of three marks. Green means you can show evidence today, such as a command output or a timestamped record. Amber means it is probably true but nobody has checked recently. Red means unknown or known to be wrong. Amber is the most common result, and it is not a failure; it is the list of things to verify next. An analytical reader will notice that this scheme mirrors the Atlas’s own staging from ad hoc through documented to baselined and measured: the further up you go, the more of the marks are green with dated proof.
Section A: Access and authority
- Who can log in remotely, and how? Evidence: the effective daemon settings printed with the validation command, not the file you remember editing.
- Are administrative rights personal? The Atlas treats a shared root-level login as the first thing to remove in a thirty-day plan. Evidence: a list of named accounts with sudo, and the sudoers.d files with correct mode.
- Do all sudo drop-in files actually load? A name containing a dot is silently ignored. Evidence: a listing of the directory plus a syntax-check result.
Section B: Time, logs and schedules
- Is the clock disciplined? Evidence: the sync tool’s status output and its configured behaviour at a leap second.
- Is log growth bounded? Evidence: explicit storage mode and size limits for the journal, instead of unexamined defaults.
- Do scheduled jobs report success? Evidence: last-run results for each job, ideally from timers with a failure hook. This maps to the scheduled-job success indicator.
Section C: Change and recovery
- Can you see what changed last? On dnf-based hosts the history can be inspected and undone; on apt-based hosts there is no undo, so the answer is a pinned earlier version. Evidence: the history output or pin list.
- How much reboot debt exists? Evidence: the reboot-required check on each machine and the date the pending restart was first flagged.
- Is a snapshot being mistaken for a backup? The Atlas states plainly that they differ. Evidence: a restore of a sample file from the backup that was actually taken, with the date of the test.
Section D: Capacity and lifecycle
- What is the disk runway in days? Evidence: the growth rate over a week and the reserved space subtracted, as in the Atlas’s worked example where a drive that looks ten days from full is really about seven.
- Is the operating system inside its support window? Evidence: an end-of-life calendar listing each host. This is the OS EOL exposure indicator.
- How long does it take to name the culprit? Evidence: the time from alert to identifying the resource at fault in the last incident. The Atlas calls this time-to-named-resource.
Section E: Safety nets on the host itself
- Is the host firewall on, and can a bad rule undo itself? Evidence: the active ruleset and a note showing that risky edits are made with a timed automatic revert.
- Is the mandatory access control layer enforcing? Evidence: its status output. Where a single service needed an exception, check that the relaxation is scoped to that one domain or profile and not to the whole machine.
- Does a reboot survive a typo in the mount table? Evidence: a verification run of the filesystem table taken before the last restart.
A quick worked scoring example
Imagine a small catalogue server. The access section scores two green and one amber, because sudo files were checked but nobody has looked at the effective remote-login settings since a migration. Time and schedules score one green and two red: no explicit log caps, and a nightly export that nobody can prove ran. Change and recovery scores one amber and two red, mostly because the only backup is a virtual machine snapshot. Capacity is amber across the board. Most of the marks are not green, and that is normal for a first pass. The useful output is the order of work: prove the export runs, take a real file-level backup and test a restore, then tidy the rest.
Reading the results with a sceptical eye
An audit is only as honest as its evidence. Three traps are worth naming. First, utilisation numbers for flash storage and RAID arrays can mislead, because the percentage of busy time does not account for parallel work; the Atlas lists trusting that figure on such devices among its anti-patterns. Second, a high load average does not prove busy processors, since the measure also counts tasks waiting uninterruptibly, often on disk. Third, a host showing plenty of unused memory in one view and little in another is usually just using the page cache; the figure to read is memory available.
A fourth trap belongs to the checklist itself. Standards documents help you decide what to compare against. The Atlas points to published server hardening guidance from the US standards institute, dating from 2008, and to the Center for Internet Security benchmarks together with an assessor tool that scores a host against them. Those are reference points for what to include; the Atlas’s scope note says choosing the baseline is a security decision handled elsewhere, while this checklist covers whether the host matches whichever baseline you picked.
Where the checklist meets the parts desk
If you buy or sell components, you already know how traceability works, and the parallels are direct. The conversation about common errors when purchasing flight control components is really about verifying provenance before money changes hands, and the explainer on how aircraft hardware is specified and sourced shows why small items deserve the same discipline as big ones. For the technology side of the business, the article on choosing business software with a structured process pairs naturally with an audit like this one.
Turning amber into green
After scoring, sort the amber and red items by how quickly a failure would hurt. For most small operations the order is recovery first, access second, capacity third, tidiness last. Pick the three highest and write each as a task with an owner and a due date. Re-run only those rows the following week, and keep the old worksheet; a series of dated audits is itself evidence of control, in the same way a stack of signed inspection sheets is.
One last point on discipline. Do not let the audit become a one-off event. The Atlas’s staged view ends with a level where the records of drills, console access tests and restores exist as standing proof. Aim there gradually. A short checklist run monthly beats an exhaustive one that nobody repeats.

