Methodology
Evidence, attribution and uncertainty
Version 0.4 · 2026-10-23
eDossier publishes dossiers about people and organisations who did not ask to be written about. That asymmetry obliges us to be explicit about how a sentence becomes a claim, what a source can and cannot prove, and what happens when we are wrong.
How verification works
- Capture. A source is recorded as an object: URL or registry reference, retrieval date, original language, translator, locator, hash, licence.
- Extract. The analyst writes claims, not prose. Each claim names its kind, its attribution, its valid time, and the exact locator in the source.
- Test. Evidence is linked with a stance — supports, contradicts, context — plus a reliability/credibility grade. Contradictions stay on the record.
- Assess. Where the record is incomplete, an assessment can be published only with its reasoning, alternatives and limits attached.
- Review. A second person checks relevance, attribution, locators and lineage, then signs a publication checklist.
- Publish and hold still. The revision is snapshotted and hashed. Later corrections create a new revision; the old one stays in the record.
Four axes that never merge
| Axis | Question | Never derived from |
|---|---|---|
| Relevance | Does this belong in a dossier about public roles, money or governance? | How interesting it is |
| Evidence quality | How reliable is the source, how credible is the information, how independent is the lineage? | How many URLs exist |
| Impact | What was observed, and what is only possible? | Confidence in the claim |
| Analytic confidence | How strong is the analyst's conclusion, given the record? | Averaged scores, virality, or how often it was repeated |
Claim kinds
- fact — a document says X, verified against the document at the cited locator.
- reported — an identifiable publication reported that someone said X. This proves the reporting, never X.
- allegation — an accusation with a named accuser or attributable publication. Never presented as adjudicated wrongdoing.
- assessment — an analytic conclusion, published with its reasoning, alternatives and limits.
Evidence grading
| Source reliability | Meaning |
|---|---|
| A | Primary, authenticated: registry, court, filing, audited statement |
| B | Primary but partial: company release, league notice, official post |
| C | Established secondary reporting with bylines and corrections policy |
| D | Unverified secondary, aggregator, or reporting with no visible method |
| E | Anonymous, partisan or otherwise unreliable without corroboration |
| U | Unknown — recorded honestly rather than guessed |
Information credibility is graded separately (1 = confirmed by other sources, 6 = cannot be judged). A grade describes the evidence, not the claim's importance and not the subject's character.
Lineage, not link counts
Translations, quotes and reposts of one article are one source. eDossier stores derived-from edges and counts root lineages, so “three publications reported it” never becomes independence unless the roots are actually distinct. Unknown dependence is not independence.
Relevance gate
Private-life material — intimate relationships, home location, identity numbers, family gossip — is excluded unless there is a documented, necessary link to governance, corporate spending, appointments or a conflict of interest. Even when public, we publish the minimum needed. Excluded material keeps only a category and a reason in our internal record.
EIDS-01 sections
| Code | Section | Status |
|---|---|---|
| 00 | Document control | core |
| 01 | Executive intelligence brief | core |
| 02 | Identity & background | core |
| 03 | Chronology & milestones | core |
| 04 | Organizational affiliations | core |
| 05 | Ownership & control | conditional |
| 06 | Capital & financial interests | conditional |
| 07 | Relationship & influence network | core |
| 08 | Significant events & controversies | conditional |
| 09 | Analytic assessment | core |
| 10 | Evidence register | core |
| 11 | Research gaps & revision log | core |
EIDS-01 is an internal editorial convention that draws on public analytic-tradecraft and investigative practice. It is not an intelligence-community standard, and no classification markings appear anywhere on this site.
Corrections and revisions
A published dossier is immutable. If a claim is wrong, we publish a new revision that supersedes it, with the previous revision still retrievable and the change logged. Removal from public view is a separate, logged decision — used for legal obligations and safety, not for tidying up an inconvenient record.
To request a correction, email hello@edossier.tech with the dossier ID, claim ID, and the source you believe we missed. We answer on the record.
What this publication is not
Not investment advice. Not a court. Not a claims-verification service for hire. Nothing here asserts that any named person has committed an offence; allegations appear only when attributable, and always labelled.