How we work

Method

Four questions decide whether a report is worth anything: how do you know, how good was the source, what else could explain it, and what would change your mind. Every report we issue answers all four on its face.

01

Everything is graded

Sources and information are graded separately, because they are different questions. A reliable source can pass on something wrong, and an unreliable source can be right. Collapsing the two into a single confidence score loses the distinction that matters most.

We use the standard scheme: source reliability A–E, information credibility 1–6. A claim graded B2 is a different proposition from one graded D4, and your report will say which is which rather than presenting both as fact.

Source reliability
GradeMeaningBasis
AReliableNo doubt about the source's authenticity, trustworthiness or competence. History of complete reliability.
BMostly reliableMinor doubts. History of mostly valid information.
CSometimes reliableDoubts. Has provided valid information in the past.
DUnreliableSignificant doubts. Has provided invalid information.
ECannot be judgedNo basis exists for evaluating the source.
Information credibility
GradeMeaningBasis
1ConfirmedCorroborated by independent sources, logical in itself, consistent with other information on the subject.
2Probably trueNot confirmed, but logical and consistent with other information.
3Possibly trueNot confirmed, reasonably logical, agrees with some other information.
4DoubtfulNot confirmed, possible but not logical, no other information on the subject.
5ImprobableNot confirmed, not logical, contradicted by other information.
6Cannot be judgedNo basis exists for evaluating the information.
02

Corroboration means independent, not repeated

Three news outlets reporting the same claim is not corroboration if all three are citing one original source. That is circular reporting, and it is the most common way a due diligence report ends up confidently wrong.

We trace claims back to origin and test whether sources are genuinely independent. Where a claim rests on a single source, the report says single-source in terms. You are entitled to know the difference between something established and something merely widely repeated.

03

Competing explanations are tested, not assumed away

The natural failure mode of any analyst is to form a view early and then read the evidence as confirming it. The defence is structured: state the plausible hypotheses — including the innocent one — and score each piece of evidence against every hypothesis rather than only the favoured one.

This surfaces the evidence that actually discriminates between explanations, which is usually a much smaller set than the total. It also makes the reasoning visible, so you can disagree with a conclusion on its merits instead of taking it on trust.

04

Probability means something specific

“Likely” is useless if the writer means 55% and the reader hears 90%. We express probability against a defined yardstick, and we state confidence separately — because a high-probability judgement built on thin evidence is a different thing from the same judgement built on strong evidence, and you need to be able to tell them apart.

≤ 5%
Remote chance
10–20%
Highly unlikely
25–35%
Unlikely
40–50%
Realistic possibility
55–75%
Likely / probably
80–90%
Highly likely
≥ 95%
Almost certain
Confidence is stated separately

Confidence reflects the strength of the evidence base and the number of assumptions in the chain. A judgement can be highly likely with low confidence — meaning we think it is probably so, but a single new piece of information could overturn it. That sentence is more useful to you than any single number.

05

AI does the work it is good at, and nothing else

We use a purpose-built intelligence platform with AI agents for collection, corroboration checking, hypothesis structuring, drafting and adversarial review. It means a single accredited assessor can deliver work that would otherwise need a team, at a price a mid-sized organisation can actually pay.

What it does not do is decide anything. The drafting agent cannot produce a sentence without attached evidence — uncited assertions are rejected and regenerated. A separate agent attacks each draft looking for unsupported claims, bias and over-confident language, and a critical finding blocks the report from being issued at all.

The key judgement is written by a human, and every report carries a named signature. That signature is the point. An assessment nobody is accountable for is not an assessment.

Where this differs from a chatbot

A general-purpose AI tool will answer any question fluently, cite nothing you can verify, and leave the entire liability with whoever passes the output on. Our constraints run the other way: nothing is asserted without a source, nothing is issued without human sign-off, and every step is recorded in an audit log that we can produce if you are ever asked to justify the decision you made on our advice.