Information security for software delivery

ISO/IEC 27001
ISO/IEC 27001 certified
The lifecycle
From a technique on paper to a rule that fires in production
A detection is not finished when it is written; it is finished when it has proven it fires, and it is never finished after that. The branch at the end is as deliberate as the start.
A flow of four stages — start from the technique, write it as code, prove it fires, watch how it behaves — with a deliberate branch for retiring a rule whose alerts stop earning triage time.
- 01
Start from the technique
Each rule begins as a hypothesis about attacker behaviour, named in ATT&CK terms so the tester, the analyst and the report all mean the same thing by it.
- 02
Write it as code
The rule lives in version control with its metadata — what it catches, why it exists, who owns it, how to triage it — and a change to it is reviewed like any change to production.
- 03
Prove it fires
Simulated telemetry that resembles the technique is replayed against the rule before deployment. A rule that has never seen the pattern it exists to catch is a hope, not a control.
- 04
Watch how it behaves
In production the question changes: do its alerts lead to analyst action, or do they train people to click acknowledge. That record decides its future.
Retire it on purpose
A rule whose alerts have stopped earning their triage time is removed through the same review path it arrived by, with the reason recorded — never muted quietly, because a muted rule looks like coverage and provides none.
The vocabulary
ATT&CK is a shared language, not a scoreboard
Mapping detections to attacker techniques earns its keep as vocabulary. When a penetration test report, a threat advisory and a detection rule all name the same technique, three teams that normally talk past each other can argue about the same thing: given this behaviour, what evidence would exist in our estate, and would any rule see it. That conversation, run technique by technique across the ones that matter to your systems, is how gaps are found while they are still cheap.
The abuse of the framework is arithmetic. Colouring a matrix and reporting a coverage percentage flatters everyone and measures nothing: a cell counts as covered whether one brittle rule watches one log source or a layered set watches them all, and the number rises fastest by writing shallow rules against easy techniques. We keep the mapping qualitative — for each technique, what would we see, on which sources, with what confidence — and we put the unwatched techniques in writing, because a gap that is named and accepted is a decision, while a gap hidden inside a percentage is a surprise on a schedule.
The artefact
What a detection carries with it into production
A rule that arrives alone is a liability; these are the companions that make it maintainable by someone who did not write it.
A version history and a named owner
The engineer who inherits the rule next year needs to know why its thresholds are what they are.
What it saves later: Changes made with understanding instead of superstition.
A test with simulated telemetry
A rule can be broken by a change to itself or by a change to the log pipeline it reads; both happen without anyone meaning to.
What it saves later: Silent breakage caught in a pipeline, not in a post-incident review.
Triage guidance attached to the alert
The analyst who sees the firing at night should not have to reconstruct the author's intent from the rule logic.
What it saves later: Consistent triage across shifts, and minutes saved on every firing.
A review date
Estates change — systems retire, platforms arrive — and a detection written for last year's estate watches a ghost.
What it saves later: A rule set that describes the estate you run now.
Our role
What we build in a detection practice and what your analysts keep
- 01
A pipeline before a pile of rules
We set up the repository, the review path, the test harness and the deployment route into your SIEM or EDR platform, so every rule after the first travels the same disciplined road.
- 02
Rules written for your estate
Detections are engineered against your log sources, your admin habits and your maintenance windows — because a rule imported from someone else's estate fires on their normal, not on your attacker.
- 03
The practice handed over, not rented
Your analysts write, test and retire detections through the same pipeline before we step back; the documentation and the tooling stay, so the discipline outlives the engagement.
Which behaviours matter most to detect, and how much alert volume your team will carry, are judgements about your risk that stay yours; we build the machinery that turns those judgements into rules you can trust and change.
Common questions
What buyers ask
What does treating detections as code change in practice?+
Three habits, none of them glamorous. Every rule change goes through review, so a well-meaning threshold tweak cannot silently blind a detection on a Friday afternoon. Every rule has a test that replays telemetry resembling the technique, so a break — in the rule or in the log pipeline feeding it — shows up in the pipeline rather than during an incident. And every rule carries metadata: what it catches, who owns it, how to triage it, when it is next reviewed. The alternative is what most estates have — hundreds of rules of unknown origin that nobody dares delete and nobody can trust.
How can a detection be tested without waiting for a real attack?+
By generating the evidence an attack would leave, safely. A technique produces telemetry — a process launch with particular arguments, an authentication pattern, a registry change — and that telemetry can be simulated or replayed against the rule without running anything harmful: crafted log events through the pipeline, benign tools that mimic the behaviour on an isolated host, or recorded events from an authorised exercise. The point is not theatrical realism; it is that the rule has demonstrably fired at least once on the pattern it exists to catch, and fires again in the test suite after every change. Purple team exercises then check the whole chain — telemetry, rule, alert, analyst — end to end.
When is deleting a detection rule the right engineering decision?+
When its record shows the alerts no longer change anyone's behaviour. A rule that fires often and is closed unread is worse than absent: it consumes triage time, and it teaches analysts that alerts from that corner of the estate mean nothing — a lesson they will apply on the day one of them is real. The decision is made on evidence, through review: what the rule was for, whether the technique still applies to the estate, whether tuning could restore its precision, and, if not, retirement with the reason recorded. What we never do is mute a rule and move on, because a muted rule sits in the console looking like coverage while providing none.
In this cluster
Review your rule set with us
A working review of your current rule set — where each rule came from, what it is mapped to, and which detections most deserve a test before anything new is written.
What happens next
- 1We reply within one business day.
- 2A 30-minute scoping call with a senior consultant.
- 3A written proposal with scope, team, and timeline.
Or reach us directly
