Methodology

Scope

What this standard governs — and the one boundary worth stating plainly: it is not how the pages you are reading were published.

Evidence classes

Three source tiers, primary preferred. Every source is fetched and read in the session that cites it, and a citation must cover the claim it sits beside, not merely its subject.

Labeling rules

Nine labels, one per assessed claim, shown beside the claim and carried wherever it is shared. Confidence — high, medium or limited — is stated, never implied.

Publication process

Five stages from fetched source to searchable record. One is a gate: a second person who did not write the assessment must approve it, and no automated identity can hold that capability.

Limitations

Where this site is below its own standard, or has no process yet, it is written down here rather than left for a reader to discover.

Corrections

An error is amended in place, marked as corrected, and announced through the channels that carried it. Nothing is quietly deleted.

This page is the standard the site holds itself to: what counts as a source, how a claim is labeled, what has to happen before anything is published, what the method cannot support, and what happens when something turns out to be wrong. It applies to every assessed claim, on every desk. Where a desk adds a rule of its own — operational reporting, the home front, influence-network research — that rule is stated in its own section below.

One boundary is worth stating plainly: that standard governs the assessment record the desk is building. It is not how the pages you are reading were published. These are edited and shipped from the project’s own repository, and the archives beneath October 7 are reproduced from their source packages as published. Where a page carries a review date and a reviewer, that is the editorial pass on the page itself.

A source is something a reader can go and check. Sources rank in three tiers, and every citation on this site names its tier by naming the publication it came from.

• Primary. The record itself — a treaty text, a government statement, a court filing, a ministry release, an original photograph or video with known provenance. Preferred wherever a primary source exists and can be reached.
• Secondary. Named reporting and named fact-checking that cites its own evidence. Used where a claim is about what was reported, or where the primary record is not publicly reachable.
• Tertiary. Reference works that summarise others’ sourcing. Used as a route to the primary documents they cite, never as the last word on a contested point.

Two rules apply at every tier. A source must be fetched and read in the session that cites it — nothing is cited from memory. And a citation must cover the specific claim it sits beside, not merely the subject the claim is about; a source that covers the topic but not the sentence is a mis-citation, and it is treated as an error to correct, not as sourcing.

A link that dies takes the evidence with it, and material of this kind is deleted more often than most. Where a source is likely to move or be taken down — a social post, a fact-check of one, a page on a site under pressure — an archive snapshot is captured at the time of citation and published beside the live link, so that the record outlives the link. The Fake Resistance case files carry those snapshots today; pages citing stable publications of record do not. An archive link is never a substitute for the primary record where one exists — it is insurance on the record that was actually used.

Every assessed item on this site carries one of nine labels. The label is shown next to the claim it describes and travels with the item wherever it is shared. No assessed claim appears without one. Each label states what the evidence has to show before it can be applied:

• Verified — supported by the evidence on record.
• False — the claim is contradicted by the evidence.
• Misleading — built on real elements arranged to create a false impression.
• Manipulated — the underlying media or record has been altered.
• Out of context — genuine material presented outside its real time, place, or meaning.
• Contested — credible sources disagree and the record does not yet settle it.
• Unsupported — we searched and found no evidence for the claim.
• Unverified — not yet assessed against the evidence.
• Satire — not a factual claim; presented as satire or parody.

A label may carry one of three confidence levels — high, medium or limited — describing how much weight the evidence behind it can take. Confidence is stated, never implied by the label alone, and a grade is never revised upward without new evidence.

Nothing reaches the public record by one person’s judgement. Five stages run between a fetched source and a searchable record, and one of them is a gate rather than a step:

1. 01

   Ingest

   A source is fetched and the fetch itself is logged, successes and failures alike, so the record of what was checked is as permanent as the record of what was found.

2. 02

   Evidence

   A claim is linked to evidence carrying its own type and strength. That evidence counts toward an assessment only once it is confirmed, not the moment it is attached.

3. 03

   Assessment

   Confidence is scored across separate dimensions rather than collapsed into one number that hides how it was reached.

4. 04

   Human review

   Gate — human only

   A second person who did not write the assessment must approve it. That capability cannot be held by an automated identity — it is refused structurally rather than by policy.

5. 05

   Publish and search

   Only what clears that review becomes part of the public, searchable record, and it carries its sources with it.

The same pipeline, and the reason the gate exists, are described on the We Are page. The boundary of what it governs is under Scope.

Reporting on the front draws on official statements cross-checked against open sources — footage, flight data, published imagery — and states only what that record supports. Where the fog is real, the reporting says so. There is no speculation about ongoing operations, and nothing that could endanger the people carrying them out.

The war is also lived far from the line: sirens and shelters, evacuated communities, the families of hostages, the slow work of rebuilding. Reporting from the home front holds to the same standard as reporting from the front — named sources, stated times — because the civilian record is the part most often distorted, and the part most worth getting right.

The case files under Fake Resistance come from a separate kind of work: open-source research into how claims move between accounts. It is worth being exact about what that method can and cannot support, because the same page argues that other people are careless with exactly these limits.

• Samples, not censuses. Posts were gathered from public mirrors and limited API pulls over stated windows. A percentage in a case file describes the sample it names, not an account’s whole output.
• Engagement figures decay. Follower counts and view counts are snapshots from the moment they were retrieved, and every one is published with that date attached.
• Video was not played. Where a claim depends on what a video actually shows rather than on its caption, the research says so instead of assuming.
• Identity is graded, and never upgraded. An account whose operator was not identified stays unresolved on the page no matter how well documented its behaviour is.
• Connection is graded too. A link is marked documented, observed, or inferred — and inference is labelled as inference rather than written up as a finding.

Findings that cut against the research’s own starting assumptions are published with the rest. Where a claim is tied to a named living person, it is because the research graded it verified at high confidence and the same conduct is already documented in mainstream reporting cited in that file.

A standard is only worth the honesty of its exceptions. These are the places where this site is below the rules above, or has no process yet. Each is stated on the page it constrains as well as here.

• One edition sits a tier below its own rule. Israel’s Story is a working edition whose chapters are largely sourced to a reference work rather than to the primary documents that work itself cites; the citation beside each claim names it, and a reader can see the tier without being told. Its ancient and biblical period is a stated gap, not an omission glossed over.
• One process does not exist yet. There is no family-consent workflow, so Our Heroes is limited to stories the subject or their family has already chosen to make public, on the record, more than once — and carries no detail beyond what is cited.
• The correction log is empty, which proves nothing. Nothing has needed a correction yet. That is a real record with nothing in it, not evidence that the standard has been tested.
• Funding is not yet public. Independence is a rule here — no sponsor gets a say in what is published or how it is assessed — but a reader cannot check it against a published list today. Support Us is where sustaining channels will be published as they open.

A network that verifies will still sometimes be wrong. The policy for what happens then, and the public log of every correction issued, are on the Corrections page.