How we source

Primary sources first

We prefer, in this order: the subject’s own words or filings; the company’s own publication; a direct interview or transcript; an established publication that names its own basis; then a reliable secondary account.

We do not use SEO content farms, quote aggregators, AI-generated summaries, or unsourced reposts. A source that is itself repeating another secondary source is recorded as secondary — it does not become independent corroboration by being republished.

Every factual line is traced

Before a post is written, its subject is researched into a structured record. Every factual sentence in the finished piece — on any channel — is mapped to a specific claim, and every claim carries its sources, how strongly they support it, and a status. A pre-publish check runs over that mapping and blocks the piece if a line cites a claim that cannot carry it.

Published posts carry no footnotes. A founder newsletter is not a paper. The traceability is internal, and it is the reason the copy can be short.

Quotes

A direct quote only appears in quotation marks if the exact wording is traceable to a source we have read. If the substance is supported but the wording is not, we paraphrase and drop the quotation marks rather than attributing specific words to a real person.

This has already changed something we published. A widely repeated line attributed to a well-known investor turned out to circulate in at least three different wordings that disagreed with each other — and with ours — and none appeared in that investor’s own published writing. We wrote that up rather than quietly fixing it.

When sources disagree

Real sources conflict. We record the conflict rather than picking the most convenient figure. Where the disagreement is material, the piece either explains the range or drops the precision entirely.

One story we published gave a specific count that circulates as three different numbers, with no primary source stating any of them. We now say that, rather than choosing one.

Numbers that change

A market capitalisation, a valuation, a headcount, a price or a spend figure is true on a date, not in general. Those claims carry the date they were true, and published copy carries it too. A figure whose date we cannot establish does not ship.

Where a source has since replaced its own figure, we say so alongside the original rather than presenting a superseded number as current.

Fact, inference and opinion

Every line is typed as one of: a fact traceable to a source, an inference drawn from named facts, editorial interpretation, or advice. An inference is never rendered as something that happened.

A great deal of what we publish is judgement — frameworks, arithmetic on worked examples, opinions about what founders should do. Those are labelled as judgement internally and are written so a reader can tell. We do not attach citations to opinion to make it look researched.

What we never publish

• Invented metrics, audience numbers or engagement figures.
• Fabricated testimonials, customers or case studies.
• Social proof we do not have.
• A download, template or resource that does not exist.

These are enforced by automated checks that run before anything can be marked ready, not by intention alone.

AI, honestly

AI tools are used in research and drafting. They are not used as a source. A model summarising something is not evidence, and nothing reaches a published claim without a traceable source behind it — which is the whole point of the process above.

Corrections

When we get something wrong, we correct it in public and leave the original where it is. A post that has been read cannot be silently edited into a different post.

The full policy, including what counts as a material correction, is in our corrections policy.