What We Will and Won’t Claim
The published method behind any claim this desk makes about a partner’s commercial outcome — the two evidence classes, the thirty-day window, and the channels no publisher can honestly see.
Independent scoring only means something if the scorer is held to it too. This desk rates vendors on what they put in writing, so this page puts our own measurement method in writing: what we count, what we refuse to count, how long a referral stays valid, and — the part almost nobody publishes — what we cannot see at all.
Why publish an attribution method at all?
Because the alternative is asking a buyer to trust a number they cannot check. Most publisher attribution is private, and privacy is doing real work there: a single figure labelled “attributed revenue” usually blends things that were measured with things that were assumed, and the blend is where the flattery hides. A brand paying for reach here deserves the same treatment a brand being scored here gets — a published method, an evidence trail, and a clearly marked boundary around what is not known. If our own numbers cannot survive being read, they should not be quoted.
What are the two evidence classes?
Every recorded transaction is assigned to exactly one class, at the moment it is recorded, and the class is stored with it. Nothing is re-graded later.
| Class | What it means | How reliable |
|---|---|---|
| Direct | The transaction carried a referral identifier this site issued, and it arrived inside the attribution window. | Deterministic — the link between the referral and the booking is observed, not inferred. |
| Partner-confirmed | An authenticated partner reports a transaction with no referral identifier attached. | Only as good as the partner's own matching. We record it as their assertion, because that is what it is. |
The two are always reported separately. There is no combined figure, and no way to produce one from what we publish — that is a property of the system, not a promise about how we will behave.
Why only two classes?
Because those are the only two we can actually observe. Longer taxonomies exist — modelled credit, fractional first-touch and last-touch splits, assisted conversions — and they are not wrong so much as chosen. A rule that hands thirty percent of the credit to the first interaction is internally consistent and externally unverifiable; it produces a number that looks like a measurement and behaves like an opinion. We would rather report a smaller set of things that are true.
What does this desk refuse to claim?
- That a booking would not have happened anyway. Attribution shows that a transaction followed a referral. It does not show causation. Proving that requires a controlled test with a genuine holdout group and enough volume to separate signal from noise — and we will say so plainly when we do not have it.
- A blended total. The classes are never added together, because the sum would be less reliable than its weakest part while looking more authoritative than its strongest.
- Modelled or fractional credit presented as measurement.
- Revenue as profit. Gross transaction value is not margin, and the two are never used interchangeably.
What can this method not see?143 words
A great deal, and it matters more than the part it can. Much business travel is booked through a corporate channel — a travel management company, a booking tool such as Concur or Navan, the GDS, a negotiated corporate rate, a phone call, or a walk-in — and none of those send anything back to a publisher. A reader can use this site, decide on a hotel, book it through their employer's system a week later, and leave no trace we could honestly claim.
The same is true of influence through an answer engine: a model can read this desk, shape a decision, and pass along no referral at all. Anyone quoting a coverage figure that ignores these channels is quoting the part of the map that happens to be lit. Our reports state the gap alongside the number, in the report itself.
How long does a referral stay valid?
Thirty days. A referral older than the window is not attributable, and a transaction that arrives after it is recorded as partner-confirmed rather than direct. The window is published here rather than held privately because an unpublished attribution window is a dial, and a dial that only one party can see is not a measurement.
What happens to a cancelled or refunded booking?
It is recorded as a reversal and stays in the ledger. Nothing is deleted. A reversed transaction is never quietly dropped to make a period look better, and reversals are reported beside completions so the reversal rate stays visible. Reported figures reflect what actually stood, not what was booked at the best moment to count it.
What is collected from readers to make this work?
As little as the method can function on. Referral links exist only on our commercial pages — the directory and the partner and Verified pages — never on articles, rankings, or scorecards. A referral record holds a random code, which partner it pointed to, the time, and which of our pages it came from: no IP address, no browser fingerprint, no cookie, no identifier of ours. No partner script, pixel, or tag runs on this site. Records expire on a schedule enforced in code rather than asserted in a policy. The full detail is in the privacy policy.
Does any of this change a score?
No, and the separation is structural rather than a matter of good intentions. Commercial measurement lives on the commercial pages; the independent rankings carry no commerce at all, and a build check refuses to publish the site if a tracked commercial link ever appears on a page that rates the company it points to. A partnership buys reach here, never a rating — and the way to verify that claim is to read the governance rules and the corrections ledger, not to take our word for it.
How does this standard change?
Materially, in public, with the date on it. If the method gains a class, changes the window, or alters what is counted, it is revised here first and the change is dated. A measurement standard that moves quietly to suit a quarter is not a standard.
The Field Brief — sharp travel-ROI notes for the people who carry the quarter through airports. No signup, no email, nothing to unsubscribe from: read it free in the open archive.
Read the brief →