The following is a fictional team story, invented to illustrate an idea; no real organisation, hospital or person is portrayed. It follows a small group in the technology office of an imagined travel-and-surgery coordination company, and it is only about record-keeping and systems administration. Nothing in it touches treatment, clinics or advice.
On a Tuesday in spring, Odalys, the office’s systems administrator, received a polite email from a partner firm asking whether the company could show which accounts had opened a shared document folder during a particular fortnight six months earlier. She said yes, and then spent two days discovering that the honest answer was “partly”. That fortnight was the subject of this entire story, and it is a decent illustration of what SecDataOps is for.
What went wrong, in order
The file server had logged access, but its default retention was ninety days. The mailbox platform had kept longer, though one of its audit categories was on a tier the company never bought. The remote-access gateway had a gap of about a day, which Ludovic, the engineer who looked after it, traced to an update that had quietly altered its logging settings. Nobody had done anything reckless. Every individual choice had been a reasonable default.
The analytical turn
Odalys wanted evidence, not blame, so she tabulated the sources. Four were silent for stretches, three had no named owner, and the company could not say how much each stream cost. She then found the Ops Atlas entry on SecDataOps, which gave her vocabulary for what she was seeing: looking after security event data so that the evidence trail is whole, believable and not wasteful. She liked its frankness about standing: the label is young and pushed by vendors, nobody in a standards committee owns it, and public bodies had written about the same chore for years under the name event logging.
The Atlas also gave her a structure. Grade every source between nought and five across five yardsticks, namely how many machines report, how many fields arrive, delay, uniformity and storage period. Write a contract for each important source. Treat parsers as code. Watch each feed for silence. Try a monthly experiment where a lab source is stopped and the alert is timed.
Measuring before changing
Her team did not rush to rebuild. They spent the first month listing sources and scoring them honestly. The totals were sobering: the average score was below two. They set retention on purpose for the top ten sources and attached a health monitor to each. Thanks to a colleague named Thaddea, a careful bookkeeper turned analyst, the cost of each stream was calculated per day, and one firewall feed turned out to cost more than all the mail logs combined.
In the second month they wrote the contracts and added test inputs for the parsers. A canary event, a tiny recurring test record, was planted in each of the ten feeds. During the third month they moved the three largest streams to a two-level arrangement, compact summaries on the fast side and raw originals in cheap long-term storage, and replayed seven days of untouched raw events to confirm nothing they relied on had stopped alerting. It followed, almost to the letter, the quarter-long outline in the Atlas.
What the numbers suggested
The story is fictional, but the arithmetic echoes a worked example in the Atlas, where a 400 GB-a-day stream costs roughly 400 units hot and about 164 when summarised and archived, with raw retention growing from ninety to 400 days. In the imagined company the saving was smaller but the principle held: longer memory did not require a larger bill.
The argument in the meeting room
Not everyone was convinced. A manager asked why the company should pay to keep old records nobody had looked at in months. Odalys answered with the table rather than a speech: the cost per stream, the age of the oldest usable record, and the single question from the partner firm that had exposed the gap. The manager did not need to be persuaded about technology; the numbers made the trade-off visible, and the budget request was approved with one amendment, a yearly review of every retention choice.
The follow-up
Six months later the partner firm asked again. This time the answer took an hour. Odalys opened the register, confirmed which sources covered the period, and pulled the raw copies from the cold tier. She also noted something the team had learned the awkward way: when a system is upgraded, the logging settings can change without anyone declaring it, which the Atlas illustrates with its own runbook example. The team added a post-upgrade check to their change process.
Lessons worth carrying
- Defaults are not decisions; someone should choose each retention period.
- An unowned feed is already failing, even if the dashboard looks green.
- Scores make arguments shorter and budgets easier to defend.
- Keep originals cheaply, since interpretation can be redone and lost data cannot.
- Test the alarm by silencing something, not by trusting that it works.
The primary reading for this approach is the Atlas page that defines SecDataOps and sets out its maturity stages, and it is worth skimming even if you only keep one idea from it.
A note for site readers
Readers who enjoy operational topics may also like our beginner-friendly look at how teams keep a fleet of working robots organised, and the broader explanation of the connected family of operations practices and how people learn them. This article is a fiction-based illustration of technology administration and offers no medical, legal or compliance advice.