Method

Two systems do the work here: a tag on every claim, and a grade on every source. Both are published with every record so the reasoning can be checked rather than trusted.

Tags: where a claim comes from

An unattributed fact is worthless, so every substantive line carries one tag.

Tag	Meaning
`STIP`	Uncontested. In a court record, agreed by both sides. Elsewhere, agreed across sources of opposing orientation
`TEST`	Sworn testimony, a statute, a regulation, an agency data release, an inspector general or audit finding
`PROS` `DEF`	Asserted by a party in a proceeding, attributed to that party
`ADV`	Asserted by an advocacy group or think tank, tagged with its orientation, never treated as neutral
`PRESS`	Reported by media without a specific underlying source
`ASSESS`	Our own analysis. Opinion, not reporting

A claim about a living person that cannot be traced to an official record or a named on-the-record source is not published as fact. It is either cut or stated explicitly as something the record does not address.

The strongest available fact is STIP across orientation. A number accepted by sources that want opposite conclusions is worth more than a number from a source you happen to trust.

Grades: how reliable a source is

Grade	Meaning
P	Primary, read in the original. The filing, dataset, statute, order or report itself
Ps	Primary, read through a summary rather than line by line
A	Staff reporting by someone present, or a named expert on the record
B	Fast reporting by someone present. Live blogs, accurate but sometimes corrected later
C	Secondary overview or aggregator, not present
D	Search-engine summary only. The underlying page was not retrieved
E	Listed as a lead, not as support

Sources that could not be retrieved are listed anyway, with the reason. A paywall, a blocked fetch, a sealed document. What is missing from a record shapes what the record can support, so hiding the gaps would defeat the point.

Limits and open questions

Every record carries two sections that most coverage leaves out.

Limits states what this record does not know: which documents were not read, which sources were not retrieved, which grades are weak.

Open questions separates what is genuinely unresolved from what has been settled and from what cannot be settled with available evidence. When a question closes, it moves, with the source that closed it.

What would settle this

Every record ends by naming the specific evidence that would resolve the question. A sealed document. A dataset an agency does not publish. A study nobody has run. If we cannot say what would change the answer, we do not have an analysis.

The freeze

The record is finished and locked before the assessment is written. Nothing is quietly added afterward because it helps the argument. Genuine later additions enter with a dated revision note that stays visible.

Machine assistance

Research here is machine-assisted. Sources are gathered and extracted with software, then graded, checked and written by a person. Every claim carries a tag showing what it rests on.

The grading system and the full source list are published precisely so the method can be audited rather than trusted. Where a source was read only in summary, its grade says so.