Methodology
How we produce a report.
Every report we publish follows the same editorial discipline. We describe the method here so you know exactly what you're reading: AI-drafted research, source-checked and signed off by a human editor, dated and cited. Reference material for informed readers - not advice for your specific situation.
Primary sources first.
We prioritise primary sources - regulator publications, statute, case law, official statistics, exchange filings, central-bank releases, and credible trade press. Secondary commentary is used to sharpen framing, not to establish fact. Every material claim carries a citation to its source, dated at point of retrieval.
AI-assisted drafting.
We use large language models (including products and suites from Anthropic, OpenAI and Perplexity) to accelerate research synthesis and structure. The model is never the author of record. It produces a structured draft against a controlled prompt taxonomy that enforces reference hygiene, currency of data, and the absence of hallucinated statistics. Every section is logged with the model version and prompt hash used to produce it.
Editorial review.
Every report passes a human editorial review before publication. Reviewers check that citations are present and verifiable, that framing is precise, and that nothing reads as opinion dressed as analysis. They are editors, not domain advisers - their job is to enforce TheRiskAgent's editorial guidelines, not to tell you what the research means for your specific situation.
When customers or our network of domain specialists flag a specific claim, the report is amended and the change is dated. That feedback loop, rather than a centrally-asserted claim of in-house domain expertise, is how we keep reports accurate over time.
Citation discipline.
If we can't cite it, we don't state it. Claims are anchored to a numbered reference list. Where we rely on estimates, ranges, or structural reasoning rather than hard data, we say so in plain text. Fabricated citations are the worst failure mode of AI-assisted research; we explicitly validate against fabrication at multiple points in the pipeline.
Dating and freshness.
Every report carries a production date on the cover. Reports in fast-moving Domains (sanctions, tech, crypto, political) are reviewed on a shorter cycle than slower-moving ones (trust law, probate). We flag when a report is stale and queue it for refresh before re-selling.
What we won't do.
- We won't publish without review.
- We won't accept paid seats inside the reports themselves. The editorial body of every report is firewalled from any firm relationship. Senior advisors can buy clearly-labelled seats whose Reports appear alongside a report on its public landing page; those are kept visually and editorially separate from the report content. Buyers pay for analysis, not for which firm is listed first.
- We won't recommend or place advisory firms or law firms inside the analysis. Where a firm's published research is cited as a source, it appears in the report's Sources section with a URL - that's evidence citation, not commercial placement.
- We won't invent statistics. If the data doesn't exist, we say so.
Corrections.
If you spot an error, write to support@theriskagent.com. Substantive corrections are made to the live report and noted with the change date. We don't silently edit.
Further reading