The grounding gate
Every scored claim carries verbatim evidence, a source, and a capture date. No quote, no claim. This rules out plausible summary that no source supports, and inference presented as observation.
How we know
Findings are only worth acting
on if they hold when disputed.
Competitive research usually fails in one specific way: it sounds right, it reads well, and nobody can trace where any particular sentence came from. Three hard controls prevent that.
The parity gate
Research depth is enforced to a comparable floor for every company in a set. Without it, the company that received the most attention scores best, and every comparison is quietly biased.
Absence discipline
If a tool returns no rows, the finding is that the tool has no data. Claims of absence require a direct check of the source, labelled as such, with a stated confidence level.
ROI construction
Six things stated before any channel is declared a winner.
Disagreements about channel performance are rarely about attribution models. They are about two teams computing different things and calling both ROI.
The conversion event
Defined precisely enough that two analysts produce the same count from the same raw data.
The revenue basis
Booked, recognised or gross profit, chosen once and stated. Channel rankings reorder depending on the answer.
The cost basis
Media plus production, agency fees and internal time. Excluding fees flatters paid channels and penalises salaried work.
The window
Matched to how your buyers actually decide, then held constant across periods.
The model
Stated plainly with its known bias acknowledged. Last click overweights branded search; first touch overweights the introduction.
The gap
The share of conversions that could not be sourced, reported as a number rather than absorbed into the winners.
Confidence labels
Every claim ships with how sure we are.
Confidence is a property of the evidence, not of the conclusion. Labelling it prevents a medium confidence inference from being quoted later as an established fact.
High confidence
Direct, dated, verbatim evidence from a primary source, checked at capture.
Medium confidence
Consistent indirect evidence, or a primary source that is dated but partial.
Low confidence
Inference from limited signal, reported as inference and excluded from scoring where it would move a tier.
When a finding is disputed
The disagreement is published, not resolved quietly.
A second, deliberately sceptical pass reads every claim against its own evidence and tries to refute it. Where a human overrules that objection, the objection stays attached to the claim with the reason.
You are not asked to trust that the disagreement was settled correctly. You are shown the disagreement, and you can look at the same evidence we did.
Flagged in review, as it appears in a report
Four case studies, all describing outcomes in general terms without specific figures.
Objection. The reviewer noted that “general terms” understates two of the four, which do carry percentage figures. Score held at tier 3 because those figures are unattributed and the tier 4 bar requires checkable outcomes.
Direct site sweep · captured 2026-08-04 · high confidence
It is the mark of a truly intelligent person to be moved by statistics.
The first principle is that you must not fool yourself, and you are the easiest person to fool.
Find out what your marketing
is actually achieving.
One conversation, no obligation: what you sell, who you lose deals to, and what you cannot currently see. If a build is a fit, it is scoped and quoted from there. There is no signup and no self-serve tier.
No prices on this site. Every engagement is scoped and quoted after the first meeting.