The audited benchmark registry

What it is

The benchmark registry is a curated set of published claims about advertising performance, each one tied to its primary source and audited against it. A typical entry is a number: a click-through rate by channel and industry, a cost per thousand by format, a typical adstock decay by channel type, the brand-versus-activation split that maximises long-run effect. Each entry carries the claim, the named source with its date, an audit status, a confidence level, and notes on where it applies and where it does not.

It is what a benchmark-led plan runs on when a brand has no history of its own.

Why it matters for a media plan

A first plan for a new brand, or a new channel, cannot come from a model, because there is nothing to fit. It has to come from what is known about brands like it. That is what benchmarks are for, and it is where most planning goes quietly wrong: the benchmark in the deck came from a blog that took it from a roundup article that took it from a slide whose source nobody can find. The number may be years old, an average of the wrong population, or made up.

A plan that states its assumptions has to be able to say where each one came from. The registry is how the Media Planner can say it.

How it works

Sourcing policy: every entry is sourced directly from an authoritative publisher: the data owner’s own report, a primary dataset’s publisher, or a platform’s documentation. Trade-press coverage is accepted only when it quotes a named report explicitly. Unsourced articles and agency blogs are banned as number sources.

Audit: each claim is re-read against its source by a reviewer who did not enter it. The audit checks that the number is what the source says, that the population and period match, and that the source is still current. Where the source gives a corrected or newer value, the entry ships the corrected value and never the original.

Status: confirmed claims ship. Updated claims ship the corrected value. Unverifiable claims, where the source could not be located or the number could not be reproduced, ship only as explicitly low-confidence rules of thumb, with the caveat inline wherever they appear. Dubious and outdated claims do not ship at all.

Confidence: high, medium or low, recorded per entry and carried into the plan. A plan built on medium-confidence benchmarks says so.

Applicability notes: most benchmarks are averages of a distribution. The entry says what the average hides: the category, the brand size, the purchase cycle, the market where the number was measured. The plan inherits the caveat.

Kuwalyst’s first registry, in 2026, was built from a widely used open-source collection of advertising rules. The audit found that it mixed reproducible data with fabricated composites, arithmetic errors, stale numbers relabelled as current, and language-model output cited as a source. The collection was rejected outright and the registry rebuilt from primary publishers, keeping only the handful of operational rules whose source was the platforms’ own documentation. That audit trail is kept as part of the registry.

How Kuwalyst uses it

Every number a benchmark-led plan uses from the registry is tagged BENCHMARK on the screen and in the export, with the entry it came from one click away. The plan lists the benchmarks it depends on among its assumptions, ranked by how much the projection would move if they were wrong, and proposes the experiment or the data connection that would replace each one with a measured or modelled value.

The Sources view of the Media Planner shows which registry entries a plan is grounded in. The planner’s review gate checks every scenario before it is shown: a figure without a tag, or a benchmark without an entry, does not pass. When the registry has a gap, the planner says so rather than filling it. The server enforces that refusal.

Over time a brand’s own evidence replaces the registry: a fitted model replaces the decay and saturation benchmarks; a geo test replaces the return benchmarks. The tags change from BENCHMARK to PROJECTED to MEASURED, and the plan shows the progression.

Pitfalls

Averages of distributions: a category benchmark is the middle of a wide spread. A brand can sit anywhere in it. Use benchmarks to size, then test.

Stale by design: media costs and rates move every year. An entry’s date is part of the entry and the plan should show it.

Precision theatre: a benchmark quoted to two decimals from a survey of a few hundred respondents is not that precise. The confidence level exists so the plan does not pretend otherwise.

The nearest number: the temptation, when the registry has no entry for a case, is to take the closest one. The registry’s rule is to report the gap. The nearest number is where invented benchmarks come from.