The story below is an invented illustration, assembled to show an idea rather than to report on any real organisation. It follows a small, unnamed operations team at a regional training provider as it learns to manage its security alarms the way an aviation safety management system manages hazards. The practice it adopts is known as DetectionOps, and the facts about it come from public references, not from the team.
The quarter that started with a clean dashboard
The team had a coverage chart that made management happy. Every cell in a grid of attacker techniques was green, because each of their alarm rules carried a label pointing at a technique in a public catalogue. The security lead, a quiet engineer who had spent earlier years in aircraft maintenance planning, asked a question that stopped the meeting: “Green according to whom, and when did we last try to break it?”
Nobody had an answer. The rules lived in a monitoring console, edited by whoever happened to be on shift. There was no list of owners, no record of why a rule existed, and no test that any of them had ever been exercised. In safety-management terms, the organisation had a hazard register with no evidence behind its entries.
Reading the reference
The lead spent an evening with the Ops Atlas entry describing DetectionOps. The Atlas frames the idea as detection engineering carried out with software delivery habits: requirements, rules as code, tests, review, automated release, measurement, tuning and retirement. It admits openly that the word itself has no formal standing, and that nearly all published guidance sits under two older labels. That candour helped, because it made the idea feel like practice rather than a product pitch.
What struck the lead most was a single worked example in which a tag-based heatmap suggested complete coverage, yet rehearsals confirmed 18 of 30 priority implementations, or 60 percent. The lesson was not that the team was failing. It was that confidence and evidence are different things, and that a safety culture exists to tell them apart.
Month one: put everything on the record
The first decision was deliberately dull. Export every rule and every exception from the console into a shared repository, one file per rule, each with a named owner and a stated purpose. Wire in a linter that bounces any file with broken structure. A teammate argued for a new commercial platform. The lead countered that the Atlas considers a tiny lab ready to start with versioned files, a pair of helper scripts and an automatic checker.
The export revealed uncomfortable facts. Two rules had no owner. Three duplicated each other. One referred to a log source the team had retired a year earlier, so it could never fire. In an aviation context this would be a finding in an audit, and the team treated it that way: logged, assigned, closed.
Month two: tests, and the trap of circular proof
Next came tests. Following guidance the Atlas relays from a vendor, each of the twenty noisiest rules received two samples, a malicious-looking event the rule has to catch and a harmless twin it has to ignore. A junior analyst proudly copied the trigger sample straight out of a rule’s own condition, and the test passed on the first run. The lead gently explained why that proved nothing. A test built from the rule passes by construction, much like checking a gauge against the same sender that drives it. They rewrote the sample from the original requirement and a captured real event, and this time one rule failed. It had a typo in a field name that had been silently excluding half the traffic.
The team also created “known gap” tests, one for each blind spot they chose to accept. When a filter removed noise from a busy rule, the filter record named its owner and an expiry date, so the trade of precision for visibility would be reviewed rather than forgotten.
An urgent change tests the process
In week eight an active campaign targeting similar organisations was reported publicly, and the team needed a new rule quickly. The old habit would have been to type it into the console. Instead they used the emergency class of change described in the Atlas: staging could be shortened, but testing and review could not, and the change would be re-reviewed within five working days. The rule shipped the same afternoon, with a short trail of evidence that the second review later confirmed.
The lead noted the difference in a hazard-report style: same urgency, but now there was a record, an owner and a rollback.
Month three: measure, rehearse and report
With rules versioned and tested, measurement became possible. The team calculated precision for each noisy rule and found one that raised dozens of alerts for a handful of genuine ones. They began tracking analyst minutes per real detection, a unit the Atlas recommends because it translates noise into human cost. They also pinned the versions of every tool in the pipeline after reading about an October 2026 episode in which a new major release of a rule-conversion library arrived while companion tools still required the older line.
Finally they ran their first rehearsal of five priority attacker behaviours and wrote a one-page report separating what was proven from what was merely mapped. Two of the five were caught. The other three became a prioritised work list.
What the review board noticed
At the end of the quarter the team presented its findings to a small review board, in the format that aviation people know well: what we found, what we changed, what remains open. The board’s questions were instructive. They did not ask how many rules there were. They asked how many had owners, how many had tests, how old the last rehearsal was, and what the plan was for the three behaviours that had slipped through. Those are the questions that move a programme forward, because each has a number or a date as its answer.
What the team took away
- A label in a coverage chart is a claim; a dated rehearsal is evidence.
- Version control and ownership come before sophistication.
- Tests must come from the requirement, not from the rule.
- Emergency changes should be fast but never unrecorded.
- Every exception needs an owner and an end date.
Why the parallel to safety management holds
A safety management system rests on four habits: identify hazards, assess risk, apply controls and monitor results. DetectionOps maps onto those habits almost line by line. Requirements identify the hazard, tests assess the control, review and pipeline apply it, and measurement monitors the outcome. The Atlas’s maturity ladder, climbing from improvised rules through versioning, testing and per-case validation to a managed portfolio, will also feel familiar to anyone who has seen a safety programme grow in stages. It adds the sensible warning, drawn from published research on a rule library, that rushing through levels leads to burnout, so the story’s team concentrated on its top threats first.
The people here are invented; the pattern is not, and any team can repeat it. Start by writing down what you have, test what matters most, and let evidence, not green squares, decide what to trust.
