Administrative teams in hospital and clinic groups buy software constantly: scheduling portals, document platforms, messaging tools, remote-access gateways. Seldom does anyone ask what the product writes to its logs, how reachable those logs are or whether they will exist when a question arises later. This list gives eight questions to put to any vendor, drawn from the practice the Ops Atlas calls SecDataOps. It concerns administrative technology only, not clinical systems or care, and it should not be read as legal or compliance advice.
1. Which events do you log, and which are optional?
Ask the vendor for the complete list of event types and a note of which ones are off by default. The Atlas gives a pointed example from cloud audit trails: object-level storage events must be switched on and are charged on top, so a customer who never enabled them has nothing to inspect afterwards. Written confirmation of what is on by default protects both sides.
2. In what format, and is there a published schema?
Records that follow a documented structure are far easier to search alongside those of other products. The Atlas names the Open Cybersecurity Schema Framework (its notes list release 1.9.0) and Elastic’s common schema as open options, with the main security-platform vendors each maintaining their own model. Ask whether the product maps to any of these, and whether the vendor warns customers before it changes field names.
3. How do I get the logs out, and how quickly?
Find out whether logs are pushed or pulled, how long delivery takes and whether failures are visible. The Atlas cautions that plain syslog cannot acknowledge receipt, so a lost message just disappears, and that a warehouse-style access view may be three hours late. Freshness targets belong in your own contract; one sample in the Atlas is a 95th-percentile delay of five minutes.
4. Are the access and audit logs part of the standard licence?
Tier differences can decide whether an investigation is possible at all. The Atlas retells how an intrusion into a government mail service in 2023 was uncovered through an access log sold at a premium level, after which a review board urged free availability with a six-month minimum. Later the vendor put that log into its standard tier, retained for half a year by default. When you buy, ask which logs sit behind which price.
5. What is the default retention, and can I change it?
Defaults are rarely tuned to your needs. Many sit at thirty or ninety days. The Atlas observes that published dwell times, a worldwide median of 14 days yet 122 in some espionage-style cases and about 400 in one campaign, make ninety days a risky default. It also flags a pricing quirk: on one cloud plan the opening month is already bundled, which means trimming retention beneath that point yields no saving. Retention duties differ by country and contract, and your adviser should confirm yours.
6. What happens to sensitive values in the log?
Logs sometimes capture usernames, email addresses or identifiers. Ask whether the product masks or hashes them, and whether you can control that. The Atlas mentions an open-source redaction processor, still labelled alpha for logs, and notes that its hashing uses a keyed method (HMAC-SHA256) because plain unkeyed hashes of low-variety values can be reversed by guessing. For anything touching personal data, involve your privacy officer before deciding.
7. What changes when you update the product?
Updates can move files, rename fields or reset logging levels. The Atlas includes a runbook example where a routine system upgrade wiped the name-lookup logging setting on two hosts. Ask for release notes that call out logging changes, and add a line to your own change checklist: “Does this change affect logs?” For a similar habit in everyday life, our checklist for first-time loan applicants shows how a short list catches omissions before they become problems.
8. Can I test that the logs arrive?
Ask how to generate a harmless test event and see it land. The Atlas proposes a monthly drill: shut a lab source down and stopwatch the wait for an alert. If the vendor offers a canary or health endpoint, even better. An untested pipeline is a belief, not a control. Record the result with a date, and repeat the exercise after every major upgrade so the evidence stays current.
Keep the vendor’s answers in writing, and file them next to the contract for that source. When staff change, the answers survive, and when a dispute arises, you can show what was promised and when.
Bonus: what to ask your own team
- Who owns the feed from this product, by name or role?
- Where is it recorded in our register of sources, and what is its coverage score?
- What will it cost per day once it is switched on, and which tier will it use?
- Are originals parked in the least expensive tier, even though we also normalise?
The Atlas’s sample contract targets, such as 99.5 per cent parse success and an alert after a quarter hour without data, are a handy reference when you write your own. They are examples, not rules. If tracking costs by hand sounds heavy, borrow the habit from our simple monthly spending routine for students: small, regular entries beat occasional heroic clean-ups.
Where this comes from
These questions are an application of the Atlas’s evidence-supply-chain model, set out in the SecDataOps entry of the Ops Atlas. The entry is upfront that the label is new, driven largely by suppliers and unowned by any standards body, and that it overlaps with what government papers call log management. Treat the list as a conversation starter for procurement, and adjust it for your own systems and legal setting.
