Editorial Standards
The standards we hold ourselves to, and what we have built to enforce them.
Our standards
Sourcing
- If we cannot source a figure, we do not publish the figure.
- Sources are graded, and primary sources win. In order: statute and SEC or FINRA rule text; an issuer's registration statement or final prospectus; exchange rule text and issuer guides; government and institutional archives; peer-reviewed research and recognised datasets; reputable financial reporting for event facts; high-quality secondary historical scholarship.
- Wikipedia and general-reference aggregators are discovery aids, never citations.
- A figure without its definition is not a fact. We state what a statistic counts and excludes, what a superlative is measured by and against, and which of several possible calculations a number uses — or we leave the number out. Traditional IPOs, SPAC IPOs, Regulation A offerings and direct listings are counted separately, never merged. A disclosed range stays a range.
- Where a claim rests on a single source, we say so. Where sources conflict, we show the conflict rather than picking a side silently.
- No rule is described as current without reading its current text.
Analysis
- We take a view. Analysis that refuses to reach a conclusion is not neutral, it is useless.
- The strongest opposing case is stated in its own best terms and then answered. A straw man is a defect, and so is false balance.
- Facts and inference are kept separate. Inference is labelled as ours, in language whose confidence is bounded by the evidence.
- We do not seek to defame. Criticism rests on what an issuer disclosed, never on imputed motive or unsourced wrongdoing.
How we enforce them
Standards that rely on remembering to apply them do not survive volume. Most of our engineering has gone into making these structural.
- Provenance travels with the fact. A published figure is bound to where it came from rather than retyped from memory further down the line.
- Verification is an ensemble, not an opinion. Facts are verified independently more than once, and agreement is required before a figure is treated as citable. Confident and wrong is the characteristic failure of language models, and one pass cannot tell the two apart.
- Numbers are mechanical; language is editorial. Deterministic checks own figures, signs and dates. Editorial judgment owns characterisation. We do not ask a language model whether a number is right, or a pattern-matcher whether a sentence is fair.
- Source grading is executed, not advised. The hierarchy above ranks the material that reaches the writer; it is not guidance the writer is trusted to follow.
- Drafts are scored against an explicit rubric — evidence, honesty, thesis, comparables, clarity, structure. Drafts below the bar do not publish.
- The system fails closed. A check that cannot be completed stops publication. An empty slot is a visible problem; an unverified number is an invisible one.
No automated system is reliable enough to trust blindly, and some errors reach publication.
Use of AI
All content on this site is AI-generated. Research, analysis and drafting are automated.
Automated analysis will contain errors — factual, interpretive, and errors of omission. Treat any specific claim as something to confirm against the underlying filing, which is the authoritative source.
Corrections
Our content may contain mistakes. If you find one, write to support@theiporadar.com with the URL and, if you have one, a source, and we will be happy to fix it.