The story below is invented, a composite sketch meant to make an idea concrete. Imagine a five-person team in Bangalore that runs an events-listing service: venues submit shows, the team curates them, and a small web app sells tickets and sends reminders. They have no security department. What they do have is a growing pile of accounts, and a Monday morning when one of those accounts misbehaved. Their response is a practical example of IAMOps, a loose label from the Ops Atlas for the steady upkeep of rules about who and what may act on which system.
Monday: the odd email
A venue manager wrote in to say that a draft listing for an unannounced show had been visible on the public page for an hour. The cause turned out to be dull. A scheduling script, built months earlier by a contractor, carried a credential that could edit every listing in the system, including drafts. A small bug made it publish something it was only supposed to read.
Nothing malicious happened. Still, the team’s lead developer drew the right conclusion: the script had far more power than its job required, and nobody had looked at that in half a year. The incident was a mild version of a pattern the Atlas describes in heavier cases, where the extent of a leaked or misused credential’s permissions decides how bad the day gets.
Tuesday: counting what existed
The team spent a morning listing every identity that could change anything: five people, two contractors, the ticketing app, the reminder service, the scheduling script and a long-forgotten staging tool. They wrote an owner beside each line. Three lines had no owner at all.
That inventory mirrors the first steps of the Atlas’s thirty-day sketch. It also gave them a surprise. One test application from a former project still held wide rights to the production database. The Atlas recounts a 2024 episode where exactly that kind of legacy test application, left with elevated access, was the entry point in a much larger organisation. The team deleted theirs by lunchtime.
Wednesday: splitting one big power into small ones
Next came the scheduling script. Instead of a credential that could edit everything, the lead created a narrow one that could read drafts and publish only listings marked approved. The ticketing app received permission to read listings and write orders, nothing else. The reminder service got read access to attendee lists and nothing more.
Anyone who has browsed pages like a rundown of theatre options around Bangalore knows how many separate hands a single evening listing passes through: producer, venue, ticketing, promoter. Each hand should see only the part of the job it needs, and the same is true for the systems behind the listing.
Thursday: a rule that blocked and a rule that didn’t
The team wanted a guardrail: nothing outside their office network may delete a listing. They wrote it as a condition on the delete action. During the sandbox test, a developer working from home deleted a test listing anyway. The reason was a classic trap from the Atlas’s list of pitfalls. The condition tested for a negated value, “not from the office network”, and when the network detail was missing from the request, the negated test behaved in a way nobody expected.
They rewrote the rule to demand a positive match and to fail closed. The lesson, which the Atlas puts bluntly, is to test guardrails before trusting them, because a rule that quietly does not apply feels exactly like a rule that works.
Friday: moving off stored keys
The deployment pipeline had been storing a long-lived cloud key as a secret. The team replaced it with a short-lived identity issued per deployment run. To avoid the over-broad trap the Atlas warns about, where a loose pattern lets every repository in an organisation through, they pinned the identity to one repository and one environment instead of a wildcard.
The same weekend, a contractor’s access lapsed on schedule instead of lingering. For a team that often talks to promoters and performers, as the ones behind a guide to music events visitors can attend in the city do, an expiring pass for outside collaborators felt natural. A day pass, not a master key.
The aftermath, in numbers
The team kept score, loosely following the measures in the Atlas. Wildcard permissions dropped sharply, stored keys went from four to one, and the average wait for a new access request stretched from a few minutes to about half a day, since a second person now approved it. Nobody complained; the extra pause was cheap next to another surprise Monday.
The Atlas itself describes a similar trade: in its worked example the time to grant a permission rose from 36 hours to 48, and the group judged that acceptable. Slower, safer, and explained, is a reasonable bargain.
Where they landed
By the Atlas’s five-level ladder, the team moved from the first rung to roughly the second: no new long-lived keys. They decided to push toward the third by routing every policy change through code review, and they planned a monthly look at who holds what. Those are modest aims, and they fit a five-person company.
Takeaways for any small listing business
- Name an owner for every account and tool. Unowned equals unwatched.
- Give each script its own narrow permission. A tool that reads should not write.
- Test every blocking rule in a safe place. Watch for missing details in requests.
- Prefer short-lived identities. Pin them to exact repositories and environments.
- Write down the numbers. A monthly count beats a vague sense of safety.
If you would like the reference material behind these habits, including the maturity model and the thirty, sixty and ninety day plans, see the Ops Atlas article on IAMOps. The Atlas admits the label is informal, which is a useful reminder that the practice matters more than the label.
For a final local touch, anyone planning an evening out on the comedy circuit, perhaps after reading an overview of comedy and live entertainment in Bangalore, is benefiting from systems like these, quietly doing their small, well-bounded jobs behind the ticket.