Evidence is rated on seven attributes, and the claim type decides which count

The rubric, the claim types, and the evidence chain

The evidence behind an association is rated on seven attributes, each 0 to 4. Directness counts double in config.yaml. Rate the evidence for the relationship the question needs, which is not the same as how much evidence exists.

Attribute	0	2	4
Directness	Nothing public addresses the question	A page covers the topic but not the finder’s problem	A page states the problem and the subject’s answer to it
Specificity	A general claim	Scope or method stated, no particulars	Named scope, method, dates, and a checkable example
Authority	An anonymous or unattributed source	A named source with limited standing	An authority for this kind of claim: the regulator, the publisher of record, the subject itself for its own facts
Independence	Only the subject speaks	One source the subject does not control	Several sources the subject does not control
Corroboration	One source	Two sources agree	Sources of different kinds agree
Freshness	Older than the claim’s useful life	Current within the claim’s useful life	Dated and recently reviewed
Consistency	Sources contradict each other	Minor differences	Every source agrees, including names, dates, and URLs

Attribute

Directness

Nothing public addresses the question

A page covers the topic but not the finder’s problem

A page states the problem and the subject’s answer to it

Specificity

A general claim

Scope or method stated, no particulars

Named scope, method, dates, and a checkable example

Authority

An anonymous or unattributed source

A named source with limited standing

An authority for this kind of claim: the regulator, the publisher of record, the subject itself for its own facts

Independence

Only the subject speaks

One source the subject does not control

Several sources the subject does not control

Corroboration

One source

Two sources agree

Sources of different kinds agree

Freshness

Older than the claim’s useful life

Current within the claim’s useful life

Dated and recently reviewed

Consistency

Sources contradict each other

Minor differences

Every source agrees, including names, dates, and URLs

Intermediate ratings (1 and 3) are for cases between the anchors. Write the source and one sentence of reasoning in note; a rating without a source is not accepted by a reviewer.

The claim type decides which attributes are required

Third-party evidence is not automatically better.

The subject’s own site is the authority for leadership, service offering, product documentation, publication metadata, and contact and location. Independence and corroboration are not_required, and they leave the average.
Independent corroboration matters most for reputation, category membership, customer outcomes, expertise, and comparative claims. Independence and corroboration are required; not_required is rejected by the validator.

Volume and purchased placement do not raise a rating

A rating cannot be raised by volume. Ten pages restating one claim are one source. A page nobody can fetch, a scanned PDF with no text, a review the subject wrote about itself, or a source the subject paid to place is not independent evidence, and the evidence interventions in interventions.md never create one.

The evidence chain has seven links, and the first weak one is where to work

For each important question, rate each link intact, weak, broken, or unknown:

finder_question -> problem -> capability -> entity -> evidence -> corroboration -> source

source asks whether the final piece of evidence is at a stable, fetchable, machine-readable address. The first broken link is the first place to work.