Research principles.

The automation is for fact-finding, not fact-shaping. This site has an obvious point of view — mass surveillance dragnets are bad, and Flock in particular has a pattern worth documenting. That point of view is allowed to decide what we go looking for. It is never allowed to decide what we say we found. A claim goes on the page because a document supports it, not because it makes a stronger case.

In practice, that means:

• Every claim traces to an actual document. A contract, a signed order, meeting minutes, a FOIA response, a news article — not a paraphrase of one. Prefer primary sources (the contract itself, the certified vote record) over secondary reporting when both exist; use news coverage to corroborate, not substitute.
• We self-host what we cite, with a link back to exactly where it came from. Official sources move or disappear; the archived copy stays checkable against the original either way, including from the citation viewer itself, so that link survives even a direct claim-level deep link that never passes through the timeline page.
• Claim-level citations link to the exact page or quote, not just the top of a file. Every quote used as a deep-link search target is verified unique in its source first.
• Unconfirmed claims are marked "Needs source," not smoothed over. A lead we haven't nailed down is a visible flag, never quiet certainty. Gaps get a note saying what's still missing, not a guess dressed up as fact.
• A negative finding is still a finding. "No Flock contract exists for this city" or "we checked and found no evidence of X" belongs on the page, documented the same way a positive finding would be — not dropped because it's less dramatic, and not invented to fill a gap where research came up empty.
• Genuine contradictions get stated as contradictions, not resolved by picking the more convenient side. When a staff report says one thing and the actual signed agreement says another, or two officials are on record saying opposite things, the entry says so plainly and lets the reader see the tension — that's a finding too, arguably the most important kind.
• No invented dates, quotes, vote counts, or attributions. If it can't be pinned to a source, it doesn't go in as settled fact — full stop, no exceptions for a claim that "sounds right."
• Every city page discloses the AI-assisted research and asks readers to email corrections. Mistakes happen; the fix is a visible, cheap path to catching them, not pretending the research is infallible because a human didn't type every sentence.

The editorializing this project does is honest about being editorializing: which cities to investigate, which documents are worth self-hosting, how much weight a finding deserves in the intro paragraph. The facts underneath aren't up for that kind of framing.

Questions or corrections

If something on the site looks wrong, or you have documents on a city we haven't covered, email contact@nosurveillancepledge.com Copy .