Data Residency: Your Backups and Logs Live Somewhere Too
Your vendor says "your data is in Türkiye" — what that sentence leaves out
In a vendor review the question comes out the same way almost every time: where is the data held? And the answer comes back the same way almost every time: in Türkiye.
Both are honest. Both are incomplete. The question asks about one tier and the answer describes one tier — the data centre where the application server runs. A whistleblowing record starts its life there. It does not stay there — and a data residency conversation that opens on that one tier usually closes on it too.
The moment a report is submitted, the same content lands in at least five other places: the overnight backup job, a disaster-recovery replica in a second region, the application logs, the error tracker that grabs a stack trace when something breaks, and the "you have a new report" email. None of these is the application server. None of them inherits the server's country automatically.
The proposal in this piece is simple. Stop asking the question once and ask it eight times. It adds one line to the conditions a channel has to meet: can the vendor document, tier by tier, which country the data sits in?
What follows is built as an audit sequence. First the question the law actually asks, then the eight tiers one at a time, then the eight questions that belong in the contract.
Moving the data is not the only route: the real question is who can read it, and from where
In practice, data residency gets argued as a matter of geography — which building the disk is in. But the risk in a whistleblowing file is not in the building. It is in the list of people who can open the file.
A hypothetical case: the server sits in a data centre in Türkiye, and so does the database. The vendor's support team works from another country and opens a remote session against the production database to look at a fault ticket. No disk moved anywhere. And the reporter's account of what happened is on a screen abroad.
So the question to put to a vendor is not "where is the server" but "who can read this data, from which country, under what authorisation". The first question asks for an address. The second asks for an access list. If nothing in your vendor file can produce the second, you do not have an answer.
The nature of the content raises the stakes. A whistleblowing record is not an ordinary category of personal data. In the official text of Law 6698, article 6(1) lists among special-category data "ceza mahkûmiyeti ve güvenlik tedbirleriyle ilgili verileri" — data relating to criminal convictions and security measures. The body of a harassment, fraud or misconduct report is exactly that in raw form: an allegation of an offence, with a name and a date on it.
The lawful bases for processing it live in the same article. Bent (d) of article 6(3) reads, verbatim: "Bir hakkın tesisi, kullanılması veya korunması için zorunlu olması" — necessity for the establishment, exercise or protection of a right. Bent (f) reads, verbatim: "İstihdam, iş sağlığı ve güvenliği, sosyal güvenlik, sosyal hizmetler ve sosyal yardım alanlarındaki hukuki yükümlülüklerin yerine getirilmesi için zorunlu olması" — necessity for meeting legal obligations in employment, occupational health and safety, social security and social services. A reporting system is grounded there, not in an employee's consent. Why that distinction matters comes up two sections down.
KVKK article 9: adequacy decisions, appropriate safeguards, incidental cases
Cross-border transfer does not run through one door. It runs through three steps, in order, and the order matters because each step only becomes available when the one before it is absent.
Step one — an adequacy decision. Article 9(1) requires a yeterlilik kararı from the Kurul covering the destination. This is not something your vendor can tell you. It is answered by looking at the decisions the Kurul has published. "That country is fine anyway" is not an answer at a contract table. Is there a decision, or is there not — that is a question you check on the day you sign.
Step two — appropriate safeguards. With no adequacy decision, transfer can proceed on one of the appropriate safeguards set out in article 9(4). The operational question here: can your vendor put in writing which safeguard it is relying on, or does it answer with "we are compliant"?
Step three — incidental cases. If neither of the first two exists, what remains is article 9(6). Its opening reads, verbatim: "...yeterlilik kararının bulunmaması ve dördüncü fıkrada öngörülen uygun güvencelerden herhangi birinin sağlanamaması durumunda, arızi olmak kaydıyla sadece aşağıdaki hallerden birinin varlığı halinde..." — where there is no adequacy decision and none of the safeguards in the fourth paragraph can be provided, then, on an incidental basis only, and only where one of the listed cases applies.
Two words in that sentence carry the whole load: arızi olmak kaydıyla — on condition of being incidental. Paragraph six is not a general route. It is a list of exceptions with that condition standing at the door.
What changed on 1 September 2024, and where explicit consent sits now
The current transfer regime arrived with Law 7499, and the date comes from Provisional Article 3. The provision reads, verbatim (Ek:2/3/2024-7499/36 md.):
"9 uncu maddenin bu maddeyi ihdas eden Kanunla değiştirilmeden önceki birinci fıkrası, maddenin yürürlüğe giren değişik haliyle birlikte 1/9/2024 tarihine kadar uygulanmaya devam olunur."
The former first paragraph of article 9 continued to apply alongside the amended version until 1 September 2024. After that date, only the new regime remains.
There is a common shortcut here and it is wrong: "explicit consent is no longer a transfer basis". The accurate version is finer. Explicit consent stopped being a primary basis. It survives as an incidental derogation under article 9(6)(a), which reads, verbatim:
"İlgili kişinin, muhtemel riskler hakkında bilgilendirilmesi kaydıyla, aktarıma açık rıza vermesi."
— the data subject giving explicit consent to the transfer, having been informed of the possible risks.
Reaching that provision means passing two gates first: no adequacy decision and none of the article 9(4) safeguards available. On top of that sits the paragraph's own condition: incidental.
Then there is the quality of that consent. The Kurul's principle decision no. 2026/921 of 29 April 2026 — its subject is the processing of biometric data for attendance tracking — reasons about the freeness of consent as follows: "çalışana rıza göstermeme veya rızasını geri çekme imkânının etkin bir biçimde sunulmaması ya da rıza göstermemenin çalışan açısından muhtemel olumsuzluklar doğurabilmesi durumunda çalışanın gerçek bir seçeneğe sahip olduğu söylenemeyeceğinden" — where an employee is not effectively given the option to withhold or withdraw consent, or where withholding it could produce adverse consequences for them, the employee cannot be said to have a genuine choice.
That decision is about biometric attendance tracking, and it lays down no general rule that employee consent is invalid across the board. But the logic in its reasoning explains why leaning on consent inside a power-imbalanced relationship is fragile. Which is why a reporting system is grounded on the bases in article 6(3) — (d) and (f), quoted above — and not on an employee ticking a box.
A continuous SaaS transfer is not "incidental"
Put the two sentences side by side.
Paragraph six opens its exceptions with arızi olmak kaydıyla — on condition of being incidental. A cloud-hosted reporting service, by definition, repeats the same flow on every report, every screen load, every backup run. That is not an event. It is an operating pattern.
A transfer that is continuous, predictable and part of the system's normal functioning is not incidental on any reading. So the consent-based bent (a) cannot be the transfer basis for a service that runs continuously. The issue is not even the quality of the consent — it is the condition standing at the door of the exception.
The practical consequence: if a vendor offers "we take explicit consent from users" as its transfer basis, the answer is coming from the wrong tier. A real answer points either to an adequacy decision or to one of the safeguards in paragraph four — or it shows that no transfer happens at all.
And "no transfer happens at all" is only an answer if it can be given for all eight tiers.
Eight tiers, eight separate questions
The table below is the checklist for your vendor file. The left column is the tier, the middle column is what data lives there, and the right column is the concrete evidence to ask for. "Yes" is not evidence.
| # | Tier | What lives there | Evidence to ask for |
|---|---|---|---|
| 1 | Application server | Raw form submissions, session data | Data centre location and region identifier, in writing |
| 2 | Database | Report text, attachments, identity fields, case history | Region, plus the list of roles with read access |
| 3 | Backups | A full copy of the database | Region of the backup store and its retention period |
| 4 | Disaster recovery | A live or lagging replica in a second region | Country of the secondary region and the failover procedure |
| 5 | Log aggregation | Query parameters, user identifiers, file names | Region of the log service and its field-masking rules |
| 6 | Error tracking | Stack traces, request bodies, sometimes form content | Region, plus proof that body capture is switched off |
| 7 | Email notification | Subject lines, summary text, sometimes the report body | Region of the sending infrastructure and what the email contains |
| 8 | Support access | Human access to production data | Countries the support team works from, plus access logs |
This table is the geographic counterpart to the question of which layer sees what. If you have asked at which layer identity becomes visible, you have to ask which country those same layers sit in. Two axes, one architecture.
In a multi-property group the table is filled in once and holds for every property — the same logic as data held in one structure across the group. If the answers differ property by property, what you have is not an architecture but an accumulation.
Tiers one and two: application server and database
These two are the only place vendors are routinely ready to answer. The answer usually arrives as a region name and is usually correct.
Two details still get skipped. First: the application server and the database do not have to sit in the same region, and in practice sometimes do not. Ask about them separately.
Second, and more important: the country the database sits in is not the same thing as the country the people who can read it sit in. That is why the evidence for tier two comes in two parts — the region, and the list of roles with read access. A vendor that cannot supply the second has not really answered with the first.
Decision rule: if the region is stated in writing and the role list is attached, the tier passes. If either is missing, the tier is open.
Tiers three and four: backups and disaster recovery
A backup is a complete copy of the database. Which means it is a byte-for-byte copy of the report narratives, the attachments and the identity fields. Writing that backup to an object store in another country moves the same content into that country as production would — only more quietly, because nobody looks at a backup on a screen.
Disaster recovery is the live version of the same thing. High-availability designs usually rest on keeping a replica in a second region, and that second region may not be in the first one's country. The question here is not "do you have backups" — everyone has backups. The question is which country the secondary copy sits in, and which country serves the data once failover happens.
These two tiers are where "our server is in Türkiye" most often breaks. The sentence is true for tier one and was never asked about tiers three and four.
Decision rule: if the country of the backup store and of the secondary region is not in writing, both tiers are open. Retention belongs in the same document — how long a copy lives matters as much as where it lives.
Tiers five and six: log aggregation and error tracking
These two barely come up in a residency conversation, and they are the pair with the highest leakage risk.
Log aggregation ships the lines an application emits to a central service. What is in those lines depends on what a developer wrote: query parameters, user identifiers, the name of an uploaded file. The filename on a whistleblowing attachment can carry meaning on its own.
Error tracking is more exposed still. Modern error trackers can capture the request body along with the exception. If an error occurs while a report is being submitted, the content of that form can land in the error record — and that record sits in the error service's country, not the application's.
So the evidence for these two tiers has to answer two questions at once: which region the service runs in, and which fields are masked. The second is shown by the configuration itself, not by a screenshot of a settings page.
Decision rule: if the vendor cannot show in writing that body capture is disabled or that fields are masked, assess these tiers on the assumption that data leaves through them.
Tiers seven and eight: email notification and support access
Tier seven is email, and it holds two separate questions. First: which country does the sending infrastructure run in? Second, and sharper: what is actually inside the email? There is a full tier of difference between "you have a new report, sign in to the console" and a message carrying a summary of the report. The second one undoes all the architectural effort inside a mailbox.
Tier eight is people. Support access is the only tier on this list that no region setting can close. If a support engineer can look at production data, the place that data lives is no longer just a disk. It is that engineer's screen.
Two documents to ask for here: which countries the support team works from, and whether every access to production data is logged. Without the second, the first cannot be audited — if nobody knows who looked and when, geography is only a statement.
Being able to answer all eight tiers from one file, out of one architectural decision, is the extension of the habit of holding property data in a single structure. Scattered tools produce scattered answers, because every tool picks its own region.
Eight questions for the contract
These are not conversation prompts. They belong in a schedule to the contract, each answered in writing and each tied to a document.
- Which country and which region do the application server and the database run in? Are they in the same region?
- Which roles can read the database, and which countries are the people holding those roles in?
- Which store, in which country, holds the backups, and how long are they retained?
- Which country holds the disaster-recovery replica, and which country serves the data after failover?
- Which region does the log aggregation service run in, and which fields are masked?
- Does the error tracker capture request bodies, and if so, which region does it write them to?
- Which infrastructure sends the notification emails, and how much of the report text do they carry?
- Which countries does the support team work from, is access to production data logged, and is that log shown to you?
One closing question follows the eight: if data leaves Türkiye at any of these tiers, what is the basis for that transfer? Data residency is only settled once that question has an answer for all eight rows. A real answer points to an adequacy decision or to one of the safeguards in article 9(4). "We take explicit consent" is not a valid answer for a continuously running service, for the reason set out in the fifth section above.
A vendor who cannot document the answer does not know the answer
The real function of these eight questions is not to collect answers. It is to make the unanswerable ones visible.
A vendor that answers tiers one and two on the spot, three and four within a day, and five and six after checking with an engineer is showing maturity — it knows its own architecture. A vendor that does not understand the question at tiers five and six is showing that those tiers were never considered. And a tier nobody has considered is a tier about which nothing can be said.
So score this on three outcomes, not two: documented, not documented, and question not understood. The third is the worst of them, because the second reveals a gap and the third reveals a blind spot.
"Our data is in Türkiye" is not a differentiator today. It is the entry ticket. More than one supplier in this market says the same sentence, and for most of them it is true as far as tier one goes. The difference shows up in the seven tiers behind it.
Walk into your next vendor meeting with this table and ask for all eight rows to be filled in. What you are looking for is a vendor that can document residency at every tier, and how many of the eight rows come back filled is the single most informative output of the whole review.
If you want to run this eight-tier check across your own vendor files, expect the table to be completed in the second round rather than the first — the first meeting usually ends with the questions being passed to an engineering team. If you are working on data residency and vendor assessment in a hotel group, we can go through these eight rows against your existing contracts together. Write to the HRMIM team through the contact page.


