How we work

A finding is an observation with its basis attached, and the decision is yours. Not "this is wrong", but "this is what we measured, and this is what it would mean."

Ask whether it was chosen

Before calling a setting a fault, we ask whether somebody chose it. That is partly so we do not say something stupid, and partly so we do not accuse anyone of something they did not do.

It is not a courtesy. In a review we ran, most of what the client pushed back on would never have been written if we had asked it first.

Somebody else checks it

When we audit an account for a client, whoever worked on that account does not grade their own work. There are three of us, so every account leaves two.

And this goes into the record the same day — who did it, who reviewed it. Separation written in afterwards is worth nothing. If an agency disputes a finding, the first thing they attack is our independence, and they never reach the finding itself.

If it does not get done, we wrote it badly

Not "the client did not follow through". If somebody tells you that you are doing ninety in a thirty zone, you do not make excuses and keep driving. If a client nods, agrees, and nothing happens, the finding did not arrive — and that is our problem to fix.

This does not weaken independence. It is what makes it sellable. "We show you what we found, the decision is yours" is independent and safe for everyone in the room at the same time.

The agency is in the conversation

Not on trial. The moment we position ourselves as the judge of an agency's work, we become a threat to the people who carry our distribution.

And the rule that makes that credible points at us: if we give a second opinion on an agency's proposal, we do not then bid for that work. Criticising something and selling the replacement is the exact conflict of interest we exist to name.

What we will not say

That the data is clean. Eighty to ninety percent is the realistic ceiling in measurement most of the time, and anyone promising more is either not measuring or not telling the truth — us included.

That we know the right answer. In this field there is no good answer that can simply be known; it depends on the company, the market and the month.

That a number is a measurement when it is an estimate. Before-and-after is not causal identification: where a reading rests on a comparison with named confounders, we say not decidable rather than producing a number.

Our own failures go out first

Nine of our own claims died under two days of research this month. Our own site failed six of the nine checks in our own audit. We publish that, because on a site whose subject is what a company tells machines about itself, our own state is either the product demonstrated or the product refuted.

The order matters: fix first, publish second. Publishing a fault while it is still live invites anyone to check it and find it still there.