What kind of document each article is
A Policy Window article is not an encyclopedia entry that summarises a subject. It is a continuously-updated scoping review built to a fixed standard, so you can re-derive it rather than trust it. This page states that standard; hold every article to it.
The standard
- 1Map every relevant instrument against the dimensions that matter.The instrument×dimension coverage matrix is the spine of every instrument and topic article — what each governance instrument does, and does not, do on each contested policy dimension. Concept and benchmark articles have no native matrix; they inherit it by linking the topics they bear on (a benchmark is a test, not a governance subject).
- 2Ground every claim in a primary source.A “governs” verdict carries the verbatim operative provision; a synthesis claim carries its citation. The verdict is re-derivable from the quoted text.
- 3Disclose both silences. Where no instrument addresses a dimension (governance silence), and — equally — where there is no replicated evidence that the governance actually works (the second silence). The absence of evidence is itself a published finding, not an omission.
- 4Answer the social-science so-what.Is the harm empirically real (problem-reality), and does governing it work (governance-efficacy)? Each is epistemic-status-labelled (established / contested / thin / absent) so a probabilistic finding never reads as settled. This is live on instrument and topic articles today; concept articles carry the section with an honest “not yet reviewed” placeholder until their evidence review is done.
- 5Breadth approaching an encyclopedia; reliability approaching a systematic review. Comprehensive coverage, with a disclosed method (see methodology) and the honest limits of a non-registered, not-externally-peer- reviewed review.
Determinate vs defeasible
An instrument article is determinate: its verdicts rest on verbatim-gateable provisions you can re-derive exactly. A topic, concept, or benchmark article is defeasible synthesis: you re-derive the substantiation and judge whether it survives adversarial review. We label which is which, and we mark AI-authored prose as a lower epistemic tier (see AI-authorship attestation).
The engine is the means, not the standard
These articles are produced by an AI “social scientist” engine. That is a feasibility bet about how to keep a comprehensive review current — not a reason to trust the result. The aim is that the engine’s involvement should be irrelevant to your confidence: re-derive us, don’t trust us. Instrument and topic articles ship with the grounding and both silences that let you do exactly that; concept and benchmark articles are being brought to the same bar (their so-what evidence review is in progress). See the cross-article evidence map for the second silence at catalog scale.