Four grades, kept apart.
Every finding carries its grade. Grades are never promoted to make a document read more confidently, and an estimate is never allowed to become a fact between one section and the next.
Observed fact
Present in official corporate records, regulatory registers or filed accounts, cited and dated. It can be looked up, and disagreed with.
Supported inference
A conclusion drawn from observed facts, with the reasoning shown and the strength of support stated separately from the facts themselves.
Estimate
A figure derived from published ranges rather than observed for this business. Always labelled, never presented as a measurement.
Unknown
Where the evidence does not reach. Reported as missing rather than filled in, and never quietly rounded into a conclusion.
Source hierarchy, in order of authority.
Where sources disagree, the higher tier wins and the disagreement is recorded. In one recent case three separate commercial data providers each reported a revenue figure that the company’s own filed accounts contradicted. All three were rejected.
Company registers, regulatory registers, filed accounts, charge registers, officer and control records.
Material the business itself publishes, read as a claim rather than as a fact.
Direct enquiry and primary research conducted for the engagement.
Licensed datasets, used for coverage and cross-checking rather than as a final authority.
Useful for events and timing. Never load-bearing on its own.
Not evidence. It can organise and draft; it cannot establish that something is true.
Six disciplines that decide what gets reported.
Compare like with like
Peer sets are built by operating model before anything is ranked. Volume placed, revenue and profit are three different measures and are never substituted for one another.
Resolve control recursively
A register names whoever was filed. Ownership is followed up the chain until the party who could actually agree a transaction is identified, or the chain is recorded as unresolved.
Absence is recorded as absence
Finding nothing is not the same as establishing that nothing exists. A sweep that returns no hits is reported as a sweep that returned no hits.
Falsify before concluding
Each surviving thesis is attacked deliberately, and so is the recommendation. What survives that attack is what gets reported.
Independent review before release
A reviewer outside the work reviews it. Where the review finds load-bearing conclusions unsound, the work goes back rather than out.
Corrections are published
When later evidence overturns an earlier finding — including one of ours — the correction is recorded in the deliverable, not silently absorbed.
What stops work from shipping.
A material regulatory or entity question that cannot be closed. A load-bearing conclusion that rests on an unverified input. An independent review that finds the central claim unsound. In each case the work goes back rather than out, and the client is told why rather than given a document that reads as though the gap were not there.
This is the part that costs time and occasionally costs a deadline. It is also the only reason a conclusion here is worth relying on.