This is an invented tale, a composite and not a report on any real firm, though every move in it is one that teams actually face. It follows a small platform crew at a made-up online freight marketplace on the night they discover someone else holds keys to their systems. The narration is deliberately plain. The aim is to show what ResponseOps, security incident response run as a standing capability, looks like when it is the difference between a bad week and a ruined quarter.
Tuesday, 21:40: the odd alert
Priya, the on-call engineer in our story, notices a service account logging in from a place no deployment ever runs. It is not dramatic. Nothing is down. Her first impulse is to disable the account and go back to her dinner. Her second, which she credits to a drill from the spring, is to ask a different question: who owns the decision to call this an incident?
The crew has a one-page plan that names an incident lead, Mateo, and a sponsor in the executive team. Priya messages Mateo not on the company chat but on a phone group they set up for exactly this purpose. The reasoning comes straight from the guidance she once read: if an intruder is inside, the team’s normal chat may be an open window.
21:55: declare, then slow down
Mateo declares the incident and writes down the time. Everyone on the call feels the pull to start pulling plugs. He holds them back, and the reasoning is worth spelling out. The Ops Atlas, in its reference page for ResponseOps, describes a myth that says you should contain everything immediately. It can tip the intruder off, and in some cases a sudden isolation sets off a prepared trap that destroys data. The Atlas does not ban fast action; it asks that any delay be time-boxed. Mateo sets a thirty-minute window for scoping before anything gets switched off.
22:10: preserve what disappears first
While the others read logs, Dalia, the platform’s most careful engineer, captures the live state of the suspicious server. She follows the order that her old training notes call the order of volatility: memory before disk, because memory is gone the moment the machine restarts. She hashes each capture as she takes it, writes the UTC time in the custody sheet and stores the files on a bucket that rejects overwrites. Nobody touches the originals again; analysis happens on copies.
It is slow work and it feels like paperwork during a crisis. Later, it is the paperwork that lets the team prove what they saw.
22:45: the scope grows
The team started with a count of six affected systems. By eleven o’clock the figure is nine, and by the small hours it is fifteen. In the Atlas’s worked example the same arithmetic gives a growth factor of two and a half, and our fictional crew finds the number uncomfortable in the same way. Every system on that list is a place where the intruder might still be, and, more to the point, every one must be dealt with before anyone attempts to throw the intruder out.
Mateo keeps a running list of persistent access: service accounts, a scheduled job, a forgotten key in a build system and a second administrator login created two days earlier. This echoes advice attributed in the Atlas to US cyber-defence guidance, which says every form of lasting access must be accounted for before the cleaning begins.
Wednesday, 03:20: eviction in one move
With the list complete, the crew prepares a single coordinated change. They will disable the rogue accounts, rotate keys, revoke tokens and cut the two network paths at the same moment. They know from the same reference that cloud rules do not always interrupt connections already open, and that short-lived credentials remain valid until they expire, so the plan includes a step for each case, not just the obvious one.
Their directory also needs a special reset. The Atlas notes that after a directory compromise, a certain master secret must be reset twice, roughly ten hours apart, because the first reset leaves the previous version working. Dalia schedules the second reset on the calendar and sets an alarm. Nobody enjoys the idea of waiting, but they agree that a half-finished eviction is worse than none.
Wednesday, 08:00: telling the rest of the company
By morning, colleagues in sales and support are asking why logins behave strangely. Mateo keeps his message short and factual: a security event is under way, the response team is handling it, here is the one number to call and here is what not to do. He does not speculate about who is behind it, a habit the Atlas backs when it observes that chasing attribution is frequently futile. The team would rather spend its hours on facts that change decisions, such as which accounts to lock and which customers to warn.
Wednesday, 14:00: recovery with caution
Restoring is tempting. The team has backups, and the temptation is to roll everything back and be done. Priya raises the question that matters: are those backups clean? Two of the systems were compromised before the last snapshot. Rolling those back would reinstall the problem, which the Atlas lists as a path to a second encryption event in ransomware cases. They find an earlier good copy for each and rebuild the affected hosts from it.
They also remember that stolen information stays stolen however neat the restore looks. That is a reminder that recovery and notification are different tracks, and that some decisions go to lawyers. At this point the incident lead brings in the company’s outside counsel to decide which reporting obligations apply and when the awareness clock began. The crew does not guess at that; none of them is qualified to, and none of this amounts to an opinion on the law.
Friday: what they changed
At the review, the crew lists five changes. First, the incident lead and the sponsor receive written authority to declare. Second, the phone group becomes permanent. Third, the custody sheet moves from a spreadsheet to a proper evidence store. Fourth, they write short playbooks for account takeover and cloud key compromise. Fifth, they schedule a tabletop with the executives for the quarter ahead.
They also start tracking a handful of numbers: time to declare, time to contain and the ratio of final scope to first estimate. The Atlas cautions that incident counts only measure workload, so they ignore the raw tally. Their new measure for the next quarter will be re-entry, whether any incident was followed by the same intruder returning.
Why tell it this way
None of the people in the story is a hero. The outcome came from a plan, a rehearsal and a willingness to slow down at the right moments. Teams that want to build the same muscle can start from the ninety-day outline in the Atlas: a lead and sponsor, a one-page plan, the separate channel and sealed emergency logins in month one; playbooks, an evidence store and a map of reporting clocks in month two; a tabletop and measures in month three.
If you are working on the design side of the system rather than the response side, our engineering blueprint for web project requirements is a useful companion, since requirements written early decide how easy a system will be to investigate later. And for teams modernising their workflows, the practical business route from paper processes toward digital platforms covers the migration habits that keep old and new systems from becoming a mystery during an incident.

