§7 revision: the LLM-prose bright line

The red-line

The current charter lists, among the §7 absolute prohibitions:

"Generating article body prose with an LLM and publishing it as a catalog row."

The 2026-Q3 revision record additionally listed the line — broadened to "no LLM-generated article body prose anywhere" — among the bright lines that revision left untouched, calling it "the structural commitment that distinguishes PW from AI-content farms". This draft proposes to withdraw that absolute formulation and replace it with §7.9 below. If adopted, the 2026-Q3 record stays as written (it documented the 2026-06-03 decision accurately); the operative rule changes here.

Proposed §7.9 — AI-assisted drafting, human-accountable publication

Article prose MAY be drafted with LLM assistance, up to and including a full machine-generated first draft, where all of the following hold:

• (a) Named-editor review. A named editor reviews the complete text against the cited primary sources before publication and approves it, taking the same per-output accountability the charter assigns for human-authored prose. The approving editor is recorded by name on the article.
• (b) Provenance disclosure.Every article carries a human-visible and machine-readable drafting- provenance disclosure (drafting mode + approving editor), extending the existing "✦ AI" labeling idiom from literature findings to article prose. Undisclosed AI drafting remains a charter breach.
• (c) Citation integrity unchanged. Every factual claim keeps its primary-source citation. The citation-integrity gate applies to AI-drafted prose identically; a claim that cannot be anchored to a cited source does not publish.
• (d) No auto-publication — unchanged. The separate bright line stands: nothing publishes to a public surface without audited human approval. §7.9 changes who may produce a first draft, not who decides publication.
• (e) Substantive-review attestation. The approving editor attests to having verified the citations and the load-bearing claims — an attestation, not a formality sign-off. The §7.2 dissent procedure applies to AI-drafted prose exactly as it applies to human-authored prose.

Operative threshold — adopted is not active

Mirroring the R5 suspension pattern from the 2026-Q3 advocacy revision, §7.9 — even if adopted — remains suspended, and the current absolute prohibition remains operative, until both of the following are verifiably true:

• the editorial board has at least 3 named editors (the same threshold that gates the §7.1 advocacy capability — one editor cannot honestly absorb AI-draft review volume); and
• the per-article provenance disclosure required by §7.9(b) — human-visible and machine-readable — has shipped in production.

Activation (not adoption) also triggers the claim-surface sweep: every public surface that states the absolute rule today — the methodology, the llms.txt family, the agent templates, the press and meta pages, and the machine endpoints — must be updated to state the §7.9 regime accurately in the same release. The inventory is tracked in the repository so the sweep ships complete, not piecemeal.

What this trades away — stated plainly

Recorded for the comment record, because a revision page that hides its costs is not worth commenting on:

• The differentiation claim weakens. "No LLM prose, period" is a category distinction; "AI-drafted, human-reviewed, disclosed" is a quality claim shared with many publishers. Per-article provenance partially compensates: a citer who relied on the absolute rule can filter to human-drafted rows.
• Review can decay toward rubber-stamping. Named-editor sign-off under volume pressure is a known failure mode. Mitigations built into §7.9: the substantive-review attestation in (e), the §7.2 dissent surface, and per-editor review-volume disclosure on the editorial-board page once §7.9 activates.
• Institutional citers may discount the catalog. Some institutions cite Policy Window because of the absolute rule. The 30-day window exists precisely so they can say so before the decision.

Why propose it at all

Editorial capacity is the binding constraint on catalog completeness — on the order of hundreds of editor-hours per year against a backlog measured in thousands (provision excerpts, deferred topic articles, companion essays, refreshes). Under the absolute rule, that scarce time goes partly to producing first drafts. §7.9 proposes converting the absolute drafting rule into an absolute review-and-disclosurerule, so human time concentrates on verification — the step that actually produces Policy Window's accuracy — rather than on first-pass prose. The catalog's typed structure, citation gates, and draft→published pipeline already exist and apply unchanged.

How to comment

During the 30-day window (2026-06-12 to 2026-07-12), comments may be submitted by:

• Email to editorial@policywindow.org with subject "Charter §7 revision comment (2026-06 LLM prose)" — the same inbox the §7.2 disagreement-resolution procedure uses.
• Public dissent under the §7.2 procedure — file the structured challenge per the steps at /wiki/charter §7.2.

All comments are logged below in "Comments received" and (where the author consents to attribution) published alongside the editorial-board response. No comment is dispositive; the editorial board makes the final decision after weighing the public record.

Comments received so far

After the window

After 2026-07-12: the editorial board weighs the public record and issues a decision — adopt, adopt-with-refinements, or decline. If adopted, the /wiki/charter §7 text is red-lined accordingly and §7.9 enters the charter in its suspended state until the operative threshold is met; this page becomes the permanent decision record either way. The 2026-Q3 revision was adopted ahead of its window by a disclosed founder decision; no such early adoption is presumed here — absent an explicit, disclosed decision to deviate, this draft runs its full window.