HRMIM
Anonymity Is an Engineering Property, Not a Promise

Anonymity Is an Engineering Property, Not a Promise

Beykan Çizel · Academic· 14 October 2026· 11 min read

Anonymous reporting is not something a piece of software promises you. It is the name of a measurement condition.

I come at this from the measurement side. A survey, a test, a reporting form — they belong to the same family of instruments, and all of them treat what a person writes about themselves as data. The oldest and most inconvenient property of that family is this: the answer is shaped by who the respondent believes will read it. Change how identifiable someone is, and you change what you are measuring. That is not a defect. It is part of the definition of the instrument.

The first half of this post deals with that measurement question — what identifiability changes, and what anonymity costs you in return. The second half is engineering: the layers a report passes through, what each layer can see, which ones we control and which one we do not. The two halves meet at the sixth heading, because on this point the reference standard agrees with the measurement specialist — and what it says will surprise most readers.

Why "completely anonymous" convinces nobody

Almost every vendor deck put in front of you carries the same sentence: "completely anonymous". It is the seller's promise, and it is a line buyers learned to discount a long time ago. I am not going to write that sentence here, because I cannot defend it — and the rest of this post shows, point by point, why it cannot be defended.

This is not a claim to be better than anyone. More than one supplier does this work, and the question is not who makes the bolder promise. It is which part of the promise can be checked.

There is a familiar version of this on the measurement side: a survey everyone was told was anonymous and nobody believed. The header says no names are collected. Then it asks for department, tenure and property, and in a small team those three fields point at one person. Your employees work that out before you do. The instrument's reliability is gone at that moment, and the only visible symptom is that the answers get softer.

On the anonymous reporting side the same story plays out at a higher price. A compromised survey produces bad data. A compromised reporting channel produces none at all.

Being identifiable changes what a person tells you

What a reporting form measures is not the incident. It is the version of the incident a person is willing to commit to writing.

If those two were the same thing, measurement would not be a discipline. They are not. The same person describes the same incident one way when they know who is on the other end, and another way when they believe the other end does not know who they are. The difference does not show up in the name field. It shows up in the level of detail: whose name is used, whether a date is given, whether the number of times it happened gets written down at all.

Three structural conditions amplify this in hotels. Teams are small and everyone knows everyone. Rota records answer the question "who was where" instantly. And in a great many roles, the person reporting works directly under the person they would be reporting.

The practical consequence: a channel's answer to "how many reports did we receive" cannot be read independently of that channel's identifiability condition. The number is not a measurement. It is a function of the conditions under which the measurement was taken.

Identifiability in the self-report literature: what gets measured changes

In self-report instruments, identifiability is not a dial. It is a condition — and when you change it you also forfeit the right to compare two measurements.

Take a hypothetical example. A group collected reports last season through a form that asked for a name, and this season through one that did not. The report count went up. In this hypothetical case the increase does not mean there are more problems in the workplace, and "more people were willing to write" is not a sufficient reading either. The construct measured in the two periods may not be the same construct. In measurement terms, comparability was never established, and the remedy is not a bigger sample. It is holding the condition still.

The same problem is well understood in psychometric instruments; what I wrote earlier about the limits of a self-report instrument applies here without modification. The instrument measures what a person is willing to write down, not what they are.

The whistleblowing literature has been working on this since the mid-1980s — the field's starting reference is Near and Miceli's 1985 paper Organizational dissidence: the case of whistle-blowing (Journal of Business Ethics, 4(1), 1-16). I am not reporting a finding from that literature here. I am telling you where to look, and nothing more. Reporting a finding requires having read the work, and writing up the findings of a paper you have not opened is one of the things that does not happen on this blog.

What to do about it: if you plan to compare report volumes across periods, record the date you changed the form's identifiability condition and put that date on the chart. If the condition changed, break the series. Do not join it up.

The measurement cost of anonymity: a report you cannot follow up is a report you cannot verify

Anonymity has a price, and you will not find it in any vendor deck.

For a report to become verifiable it usually needs a second round: which date, how many people were present, do you still have that message. If you cannot reach the reporter, there is no second round. You are left with a single-pass narrative, and a single-pass narrative is often enough to open an investigation but rarely enough to close one.

The reverse holds too. A channel that captures identity in order to improve follow-up has changed the condition described two headings ago, and the report count will fall. You do not have to pick one end of that trade-off. Believing you are not picking is the expensive mistake.

This is where measuring the thing you actually mean to measure comes in. If report volume is going to be used as an indicator, write down in advance what it represents: the accessibility of the channel, the density of problems in the workplace, or the trust people place in the channel. One number cannot carry all three.

What to do about it: leave a route back to the reporter without making identity mandatory. The tenth heading covers how that is built. The point here is that this is solvable without giving up anonymity.

Confidential and anonymous are not the same thing

These are the two most frequently conflated words in this market, and the conflation is not accidental — one of them is much easier to deliver than the other, and both get sold under the same label.

The distinction is simple. Anonymity means the system never held the identity in the first place. Confidentiality means the system holds it and restricts who can see it. In the first case there is nothing to protect; in the second there is, and protecting it depends entirely on an authorisation design. That difference — concealing an employee's identity versus never collecting it — is an architectural decision, not a contract clause.

Directive (EU) 2019/1937 asks for the second of the two. Article 16(1) reads:

"Member States shall ensure that the identity of the reporting person is not disclosed to anyone beyond the authorised staff members competent to receive or follow up on reports, without the explicit consent of that person. This shall also apply to any other information from which the identity of the reporting person may be directly or indirectly deduced."

Scope caveat, and it applies to every EU citation in this post: this is EU law. It does not bind a privately held Turkish hotel group. I am using it as a reference standard, because an investor, a chain partner or an acquirer is likely to read you against it — never in the sense that the law requires this of you.

And the second paragraph of the same article shows why an absolute confidentiality guarantee cannot be given:

"By way of derogation from paragraph 1, the identity of the reporting person and any other information referred to in paragraph 1 may be disclosed only where this is a necessary and proportionate obligation imposed by Union or national law in the context of investigations by national authorities or judicial proceedings…"

So the standard does not say identity will never be disclosed. It leaves an express carve-out for investigations by national authorities and for judicial proceedings. Any page promising that a reporter's identity will never be revealed under any circumstances contradicts the very source it is leaning on.

Which one the reference standard actually requires — and the answer is a surprise

Most readers assume the EU standard mandates anonymous reporting. It does not.

Article 6(2) reads:

"Without prejudice to existing obligations to provide for anonymous reporting by virtue of Union law, this Directive does not affect the power of Member States to decide whether legal entities in the private or public sector and competent authorities are required to accept and follow up on anonymous reports of breaches."

The decision is left to member states. The same logic shows up in Article 9(1)(e), where diligent follow-up on anonymous reports applies "where provided for in national law" — that is, only to the extent national law provides for it.

What the Directive does require is confidentiality. Article 16, which you read a moment ago, says "shall ensure", not "may ensure". Anonymity is an option. Confidentiality is a condition.

For anyone who works in measurement, this is not surprising at all. Anonymity is an extreme case; confidentiality is a design property that can be specified, graded and inspected. The standard made mandatory the thing that can be delivered and left open the thing whose delivery cannot be guaranteed. That is why the right question in a vendor evaluation is not "is it anonymous". It is: is visibility of identity structurally constrained, and is that constraint enforced by the system or only by a policy document.

The narrative identifies the reporter on its own

The most important part of Article 16(1) has nothing to do with the name field: "any other information from which the identity of the reporting person may be directly or indirectly deduced". Any information from which identity can be indirectly inferred.

This is not a problem you solve with database columns. An example — hypothetical, not drawn from any real incident: in a 40-room boutique property, the phrase "the night manager on the second Saturday of August" describes exactly one person. There is no name field on the form, no IP in the record, no session identifier. The narrative has still said who it is about. In this hypothetical case, what exposes the reporter is not the system. It is the shape of the event.

There is no fix for that. There is a way to manage it, and all of it runs through being straight with people.

What to do about it: say it to the reporter, on the form, in plain words. "The details you write may identify you; you decide how much detail to give" is both more accurate and more protective than the promise this post opened by refusing. That promise encourages a reporter who believes they are safe to write more detail. This sentence hands them the decision.

Layer by layer: who sees the form, who sees the record, who sees the infrastructure

The only honest way to discuss anonymity as a property is to count the layers a report passes through. There are four, and each one sees something different.

The form layer. The fields the reporter fills in. Control here is complete: which fields are asked for, which are mandatory, and which are never asked at all are design decisions, made by you.

The record layer. The report as it exists in the database, and who can reach it. We set out the conditions a reporting channel has to meet in an earlier post, and the first of them was exactly this: preventing unauthorised access by construction — through authorisation, not through policy. A shared mailbox cannot satisfy that condition by definition.

There is a party most discussions of the record layer skip entirely: the person the report is about. Article 6(1) of Turkish Law 6698 on the Protection of Personal Data, when it enumerates special categories of personal data, includes "ceza mahkûmiyeti ve güvenlik tedbirleriyle ilgili verileri" — data relating to criminal convictions and security measures. A report narrative alleging an offence therefore produces special-category personal data not only about the reporter but about the person accused. The article in its current form was re-enacted by Law No. 7499 (Kanun No. 7499, 2/3/2024, m. 33).

The infrastructure layer. The servers the data sits on, the backups, the logs. There is control here, but it has to be technical and it has to be documentable.

The edge layer. And here there is no control. The next heading is entirely about that.

The IP address: the edge sees it, we do not store it

Every request that reaches a web form passes through an edge network before it reaches the application. That network sees the IP address the request came from. It is not a preference. It is a consequence of how the connection is made.

I am not going to tell you we can change that, because we cannot. The claim we can make is narrower, and it is exactly this: we do not store the IP address with the report. We cannot stop it being seen. We can choose not to write it down, and that is the choice we made.

This paragraph is short on purpose. For a sceptical buyer, naming the layer you do not control and drawing the boundary out loud is worth more than three pages of security assurances. If a supplier tells you the edge does not see the IP address, your next question should be "how".

The return channel: how you reply to someone who stayed anonymous

The problem left open four headings ago was this: if you cannot reach the reporter, you cannot verify the report.

The way through it is to separate identity from the access key. A tracking key generated when the report is created, held only by the reporter, makes a two-way conversation possible without the organisation ever knowing who they are. What the organisation sees is a case number. What the reporter sees is their own case.

There is a cost to this and it should not be hidden. A reporter who loses the key cannot get back into their case, and the organisation cannot identify them and hand it back — if it could, there was no anonymity to begin with. That follows from the definition of the system, not from a flaw in it.

What to do about it: on the screen where the key is shown, state on the same screen what happens if it is lost. A reporter making that decision knowingly is better than an explanation delivered afterwards.

The reporter identified later: the standard assumes this happens

The most honest provision in the Directive is the one least often quoted. Article 6(3):

"Persons who reported or publicly disclosed information on breaches anonymously, but who are subsequently identified and suffer retaliation, shall nonetheless qualify for the protection provided for under Chapter VI, provided that they meet the conditions laid down in paragraph 1."

The existence of that provision is itself the finding. The people who wrote the standard assumed that reporters who filed anonymously would subsequently be identified, and built the protection around that assumption. They did not assume anonymity would hold and hang the protection on it.

The scope caveat applies here too — this is EU law and it does not bind a privately held Turkish hotel group. The design lesson transfers intact, and it is this: retaliation protection must not depend on anonymity holding. Write down what you will do for a reporter whose identity becomes known before their identity becomes known. In a seasonal operation, that means having a record of who took the contract-renewal decision and on what grounds.

What can honestly be promised, and what cannot

To close. Anonymity is not one property. It is several, and each of them is either engineered or it is not.

What can honestly be promised:

  • The form does not ask for identifying fields, and a field that is never asked for is never stored.
  • The IP address is not stored with the report.
  • Access to the record is constrained by authorisation, and that constraint is enforced by the system rather than by a policy.
  • A reporter can hold a two-way conversation without giving a name.
  • The reporter is told plainly that the narrative itself may identify them.

What cannot be promised — and leaving this list out is what devalues the first one:

  • That the edge network does not see the IP address. It does.
  • That the narrative will identify nobody. In a small property, one sentence is enough.
  • That identity will never be disclosed under any circumstance. There is a carve-out for legal process, and the reference standard writes it out expressly.
  • That identity will never come to light later. The standard itself is built around the assumption that it will.

The second list does not weaken the first. It is the only thing that makes the first believable. A compliance officer has already stopped accepting promises whose limits are unwritten.

If you are running a vendor evaluation, turn both lists into your own questions and put the second list to the other side. The supplier who hands you the second list before you ask is describing a system that engineers anonymity as a property; the one who does not has left the first list on paper.

If you want to map what each layer of your own reporting channel can see — form, record, infrastructure and edge, separately — we can work through that assessment with you. Get in touch, and let's start from which conditions your current setup already meets.

Get New Posts by Email

We will email you when a new blog post is published.

Loading the form…

Related Posts

Data Residency: Your Backups and Logs Live Somewhere Too

Data Residency: Your Backups and Logs Live Somewhere Too

When a vendor says "your data is in Türkiye", they are usually describing one tier: the application server. A whistleblowing record also lands in backups, a disaster-recovery replica, logs, an error tracker, a notification email and a support engineer's screen. This piece walks the three steps of KVKK article 9, explains where explicit consent sits after 1 September 2024, and puts eight tiers into one table with the evidence to demand at each.

28 October 2026 · 11 min read

Budget vs Actual: Does the Variance Measure Performance or the Budget?

Budget vs Actual: Does the Variance Measure Performance or the Budget?

When a line closes twenty percent under budget, the first instinct is to question the operation. But a variance is the difference between two things, and one of them is the budget itself. This piece treats the variance as a measurement problem: is it measuring performance, or the accuracy of the budget? The literature on budget-based evaluation carries a clear warning here — and the contradiction inside that literature shows how much weight one variance figure can carry.

4 November 2026 · 13 min read

Off the Payroll: Who Can Actually Use Your Ethics Hotline

Off the Payroll: Who Can Actually Use Your Ethics Hotline

Laundry, security, landscaping and spa are rarely on a hotel's payroll — but they work the same corridors. Limiting a reporting channel to payrolled staff is usually a habit, not a decision. This post frames the scope question with two reference points: joint principal–subcontractor liability under Labour Act art. 2, and the reporting persons listed in Directive art. 4. Then three costs of narrow scope, two of wide scope, and a four-question frame for writing the decision down.

30 September 2026 · 11 min read

Let's work with HRMIM

Explore our platform free of charge.