The machine can fetch it. Only a human grades it.

1 · No lying. No invented quotes, sources, numbers, or events — ever. A blank beats a fabrication.
2 · Triangulate, don't inflate. A claim needs more than one independent source before it's stated as fact.
3 · Primary over secondary. Link the document, the dataset, the on-the-record source — not a hot take about it.
4 · Label the inference. Anything the machine reasoned rather than sourced is marked as inference until a human verifies it (the house Inference Clause).
5 · Separate fact from framing. What happened, then who's spinning it which way — clearly split.
6 · No manufactured outrage. The desk grades down for stories built to enrage rather than inform.
7 · A human signs every grade. The machine drafts the grade; a human ratifies or overrides it. Only humans score.

"Auto-imports and researches it the way I like" needs three things this static, locked-down museum doesn't have yet: a backend job to pull feeds on a schedule, the owner's API keys for the research model, and a relaxed network policy — the strict CSP here (connect-src 'self') deliberately blocks all outside fetches, which is a security feature, not a bug. So a real auto-desk is a build-pipeline / worker job, not a client-side script.

The honest path: a scheduled research worker drafts graded items into the site (like the daily accessibility audit already does), then a human ratifies each grade before it publishes. Until that's wired with the owner's keys, no auto-imported headlines appear here — because faking them would break clause #1.

The live grading desk — where the curator pulls, grades, and signs — is meant to be the owner's, gated behind the login the launch plan already scaffolds, and likely lives on declassifiedbysean.com. A public static page can't truly enforce "owner only" (no real auth client-side), so this page publishes the method openly while the private, auto-fed desk waits on the backend + login. The standard is public; the desk is the curator's.