Our client was winning every category. That was a bug.

The first live competitive audit came back with the client leading every single category.

On most engagements that is the slide you open with.

Why it was wrong

The scoring standard requires three independent, dated, verbatim pieces of evidence to award tier 4 or tier 5. That rule exists to stop a strong impression becoming a strong score. It works, and it has a consequence that is not obvious until it bites.

We had gathered 17 pieces of evidence on the client and an average of 6.5 on each competitor.

At 6.5 records, a competitor cannot reach a top tier on any lens. Not because they do not deserve it. Because the proof required to award it was never collected. They were structurally capped.

So the grid was not measuring the market. It was measuring how hard we had looked at each company, and we had looked hardest at the one paying for the report.

Nothing in the output revealed this. Every individual score was correctly awarded against the published requirements. Every piece of evidence was real, sourced and dated. The grounding gate had rejected nothing except its planted controls. Each cell was defensible in isolation.

The failure lived entirely in the distribution, and no individual record could show it.

Why it happens by default

This is not a mistake anybody has to make. It is what happens when nobody prevents it.

You have privileged access to your client. You have their analytics, their team, their internal context, their time. Everything is easier to find and there is more of it.

Competitors are researched from the outside, at a distance, in less time. That asymmetry is not laziness. It is the natural gradient of the work.

Add an evidence floor to your scoring, which any rigorous standard needs, and the gradient silently converts into a scoring advantage for whoever you had the most access to.

Every competitive audit built by a party with a client relationship has this failure mode available to it. Most, I suspect, have it.

The fix

We went back and researched the entire field to the same depth.

293 evidence records across the cohort. An average of 26 per competitor. None below 18.

The client dropped from first place to second.

We shipped that.

What that cost and what it bought

The immediate cost is obvious: a considerably less pleasant conversation, and a report that says a competitor is ahead.

What it bought is the only thing that makes any of the other numbers worth reading. If the grid can only produce a result the client likes, the grid is a compliment with axes on it. The client knows that, their board knows that, and any competitor shown the document knows it.

A scorecard that is capable of reporting bad news is the only kind whose good news means anything. Those are the same property, and you do not get to have one of them.

The gate

Fixing the instance is not fixing the problem, so the finding became a control.

The parity gate measures evidence depth across the cohort automatically, before anything can publish as final. It checks four things:

• Total evidence records per company
• The ratio between the most and least researched company
• A hard floor below which no company may sit
• Whether any tier 4 or 5 award rests on fewer than three independent dated records

Until all four pass, the report cannot publish as final and the dashboard says so in plain language, naming which company is under-researched.

It is a gate rather than a warning because of how this failure presents. A warning is a thing somebody clicks past. And the failure it prevents is invisible in the output, which means nobody downstream will catch it. Controls for invisible failures have to block.

The rule that came out of it

A client score that beats the field must be beatable by the field.

If your client’s win could not have gone the other way given the research you did, you have not measured anything.

And the operational version, which now runs on every audit:

A client leading every category is a red flag to investigate, not a result to report.

That is now the first thing checked when a grid comes back looking excellent. Not because a client can never lead everything, but because the probability that a real company genuinely leads five independent lenses against five to ten competitors is low, and the probability that uneven research produced it is high.

An unusually good result is a hypothesis about your process before it is a fact about the world.

Why this is on a public page

Because it is the best evidence we have for anything else claimed here.

We could describe the parity gate as a feature and never mention what produced it. It would sound like foresight. Instead: we built something, it produced a flattering and wrong answer, we caught it, we fixed it at real cost, and we shipped a worse-looking number for our own client.

That story is checkable in a way a feature list is not. The numbers are specific. 17 against 6.5. Then 293, mean 26, minimum 18. First place to second.

A product about honest measurement that could not describe its own most embarrassing failure would be worth exactly nothing.

The parity gate, and the three other controls it sits alongside, are described in full on how we prove it.