Chronolith by Aperlock · Capability brief

The Problems Nobody Alerted On

A firm without an IT department does not need more alerts. It needs a short, calm list of things worth fixing this week — the credential nobody retired, the server that stopped reporting, the disk heading for a wall.

Capability Operations intelligence Standing checks 45 Cadence Every four hours, weekly digest Updated 7 September 2026

In short

  • Findings, not alerts. These are “worth fixing this week”, deliberately kept out of the urgent channel so they cannot cause fatigue.
  • 45 standing checks across account hygiene, system health, exposure and behavioural drift.
  • They clear themselves. Fix the underlying condition and the finding resolves without anyone closing a ticket.
  • One finding per problem, not one per event. A credential retrying every twenty minutes is one item with a host list, not a thousand rows.
  • A weekly digest at 9am Monday, written to be read over coffee rather than at 2am.
The problem

Most estate problems are not alerts

An account disabled two years ago that still tries to authenticate every twenty minutes from three machines. A domain controller that stopped reporting eleven days ago. Local administrator accounts that appeared on three workstations last month. A disk that will be full in six weeks.

None of these is an emergency. All of them are how an incident starts, and none will ever be noticed by a firm whose IT is a part-time arrangement — because nothing is alerting.

Why they cannot simply become alerts

Route these through the urgent channel and you get alert fatigue, which reliably ends with the urgent channel being ignored too. These conditions are monitored; what they need is a different tempo.

The model

Two channels, two tempos

Alerts Acute and urgent. Something is happening now. Delivered immediately to wherever you want to be interrupted.
Findings Standing conditions worth fixing. Evaluated every four hours, shown on the dashboard, summarised weekly.

The separation is enforced in the language too: findings are never called alerts, never called critical, and never phrased urgently. A calm channel that starts shouting stops being a calm channel.

Findings resolve themselves

A finding is a standing statement about the estate, not an event. When the condition clears, the finding closes automatically. There is no queue to groom and nothing goes stale because somebody forgot to tick it off — which is exactly what happens to a ticket list at a firm with no operations team.

Coverage

What the 45 checks look for

What this looks like in practice

A terminal server where the remote-desktop service and a single sign-on agent were both failing several times a day. Each time, everyone working on that server lost their session. Each time, the services restarted themselves within seconds and the server never went down — so uptime monitoring reported a healthy estate throughout, for weeks.

Chronolith reports it as one finding rather than as an alert per crash: which services, how often, on how many machines, how close together, and which one failed first. It also shows how many of each service’s failures had the other service anywhere near them — because on that estate one of them failed fifteen times and only four of those coincided, which meant the explanation everybody had been working from was the wrong way round.

Finding categories, with representative examples
CategoryExamples
Account hygiene Disabled accounts still authenticating; a disabled account that succeeded; dormant accounts; passwords past their age; new local administrators; privileged group changes; service accounts logging on interactively
System health Sources that went silent or stalled; forwarders offline, unable to read one channel, or reading fine and unable to deliver; audit policy switched off; security tooling in maintenance mode; clock drift; permission changes not being recorded; permission data that has aged out or was never collected; services that keep failing and restarting while the server itself never goes down; the product's own background components failing, or silently not running at all
Exposure Remote desktop reachable externally; unauthorised remote-access tools; new unregistered hosts; new services; break-glass account use; no visibility of data leaving the estate; file-sharing protocols reaching the internet; devices on the network that report nothing about themselves
Drift Logon volume shifts; first-time administrator behaviour; access boundaries being crossed; detections that have stopped firing; patch and backup gaps
Some need time before they can say anything

Dormancy, patch gaps and backup gaps all require a baseline — they are statements about something that used to happen and stopped. Those take thirty to sixty days of history before they mean anything, and the product says so rather than reporting a clean result from an empty baseline.

Design

One finding per problem, not one per event

A retired service credential still held by a scheduled task can produce thousands of failed authentications before anybody notices. Alerting on each one is an alert-fatigue machine.

The value is the host list and the cadence

“Retries every 20 minutes from these three machines” is a fix path. A thousand rows is a scrollbar.

These ghosts survive for years precisely because nobody knows what still depends on the credential, so every attempt to clean it up breaks something. Naming the machines that still hold it is what makes the cleanup possible.

The same principle applies throughout: a finding names the affected set and what to do, rather than reproducing the evidence that produced it. The evidence stays searchable.

Capacity

Why it will tell you to grow but rarely to shrink

Capacity advice is where monitoring tools most often embarrass themselves, and the reason is a confidence asymmetry that most of them ignore.

“This needs to be bigger” A present-tense fact. If something is at 90% now, a week of data proves it. Reported immediately.
“This could be smaller” A claim about a peak not yet observed. Absence of evidence is doing all the work.
A finance server at year-end

A finance server sits at four processors for fifty weeks of the year and is increased to twenty-four for the two weeks of year-end. Sample it in June and every statistic says shrink it. Act on that and the firm discovers the mistake at the worst possible moment.

So redlining — a present-tense fact — reports immediately with no history required. Headroom trends report with the window they were measured over stated alongside, so nobody mistakes thirty days for a year.

Delivery

A Monday morning email, not a pager

  • Grouped by category, so related work is done together rather than one item at a time.
  • Formatted for action — what the condition is, which machines or accounts it affects, and what to do.
  • Weekly by default, changeable to monthly or off without editing configuration files.
  • Read at 9am Monday, not at 2am. The tone is deliberate and so is the schedule.
Who this is written for

Not a security operations centre. A managing partner, an office manager, or an external consultant who visits monthly. The digest has to make sense to somebody who was not watching the console all week, and has to be short enough that they read all of it.

Questions we get

Common questions

Is this just a compliance checklist?

No. Every finding comes from telemetry the product already collects about your actual estate, not from a questionnaire. A checklist asks whether you have a policy; this reports that a specific disabled account is still authenticating from four named machines.

How do we stop it becoming another ignored queue?

Findings resolve themselves when the condition clears, so the list shrinks as work gets done rather than needing to be groomed. And the whole point of separating them from alerts is that this list is short enough to read — if it were not, it would be alerts again.

Can we dismiss findings we do not care about?

Findings can be acknowledged, and suppression exists with an expiry so a decision to ignore something is revisited rather than permanent. A silenced check that stays silenced for two years is indistinguishable from a check that was never written.

What if we have no baseline yet?

Several checks need thirty to sixty days before they can say anything, and they report that they are still establishing a baseline rather than reporting a clean estate. An empty result from an empty baseline is not a clean result, and the product does not present it as one.

Does this replace an IT provider?

No, and it is not meant to. It tells you what needs attention in language you can hand to whoever does the work — internal or external. For firms with an external provider, it is often most useful as the agenda for the monthly visit.

All capability briefs Get Chronolith →