Methodology.

A field guide, run on a budget.

Anomaly Daily covers UAP, ancient mysteries, cryptids, paranormal incidents, unexplained historical events, and weird weather. We treat the subject matter as interesting and the evidence as testable. We are neither a debunking site nor a true-believer site — “unconfirmed” is a perfectly good ending, and most cases end there.

The site ships a daily puzzle, an interactive map, a live signal pulse at /live, and a growing archive of source-cited cases. One person operates it, with help from the AI tooling described below. We are explicit about which parts of the work humans do and which parts machines do.

Every claim earns a citation.

Every case page lists the sources it's built on. We prioritize, in order:

1. Official documents.Declassified files, FOIA returns, the digitized Project Blue Book corpus, the U.S. Department of War's PURSUE releases, congressional testimony.
2. Primary witness accounts. On-the-record statements from named witnesses with verifiable affiliations — pilots, navigators, site staff, named police reports.
3. Academic + scientific work. Peer-reviewed papers, journal articles, university and lab reports.
4. Journalism of record. Reporting from outlets that carry named bylines and editorial standards — the NYT, Reuters, the AP, Defense News, etc.
5. Community analysis — Metabunk, Enigma Labs, and individual independent researchers — used and explicitly labeled when the analysis is load-bearing for a case.

We do not source from anonymous social posts, unverifiable YouTube channels, or tabloid aggregators. Witness names appear in case copy only when they appear in the cited primary record.

How /live decides what to show.

The live page shows clustered stories, not raw signals. The pipeline runs hourly on the operator's machine, polling ten public sources — Reddit, X, Bluesky, YouTube, Google Trends, the NUFORC report stream, Wikipedia's recent UAP edits, war.gov, curated Substacks, plus Polymarket and Kalshi market readings. Each signal is scored by a local language model on relevance, velocity, and novelty, then tagged with one of the brand's six topics. Signals the classifier can't confidently place stay out.

Same-topic signals about the same underlying event collapse into a story. A story is what readers see; a single Phoenix-triangle sighting that hits five sources becomes one entry on /live, not five.

A story appears on /live only if all of the following hold:

• It's in one of the brand's six topics.
• Its aggregate score crosses 600 / 1000 (a “warm” floor).
• At least two distinct sources corroborate it — or one source carries a strong-enough individual signal (≥800).
• The operator hasn't dismissed it.
• It has new activity in the last 24 hours.

The operator can also promote a story manually — on slow news days, a Pentagon memo deserving the spotlight gets pinned regardless of corroboration math — and can dismissa story that looks redundant, off-topic, or inappropriate. Both actions affect /live immediately. The dismiss log persists; the same false-positive doesn't resurface tomorrow.

One number for “how anomalous is today?”

The Index is a single 0-100 reading computed once a day from the day's signal volume, peak engagement, novelty against a trailing-30-day baseline, geographic spread, and the variance in our curated prediction markets. The five components are z-scored, weighted, run through a sigmoid for the 0-100 bound, and smoothed with a seven-day exponential moving average so single-day spikes don't feel chaotic.

The Index is calibrated against itself — a 73 means “73 vs. recent baseline,” not “73% of cosmic truth.” We surface the number on /live during the validation period; a fuller methodology page with named components and a downloadable history lands once we have enough days logged to evaluate stability against external events.

What the markets price, in their own words.

The markets widget on /live shows current Polymarket and Kalshi prices on a curated list of UAP-relevant contracts — disclosure timelines, alien-confirmation deadlines, file-release expectations. We don't price the markets; we report what the markets price. Tickers are kept in data/markets/curated.json and reviewed when new derivative markets open.

CTAs to view a market on its platform carry an analytics tag that lets us measure referral volume; if any platform offers us a paid affiliate or sponsorship relationship, we'll disclose it directly under the relevant market and on this page. As of this writing we have no commercial relationships with Polymarket or Kalshi.

We surface verdicts. We don't arbitrate.

Where an authoritative body has issued a verdict on a case — the Air Force on Phoenix Lights, UNESCO on the pyramids, NOAA on a weather phenomenon, AARO on a 21st-century UAP — we surface their classification verbatim and dated. When verdicts conflict, and they often do, we show both. We do not declare which agency was right; that's the reader's job.

Machines carry the volume. Humans carry the judgment.

AI is load-bearing here. We say so on the front of the page rather than in a footer.

• Drafting. Puzzle drafts, case summaries, and social copy variants are drafted by language models — Claude, GPT-class models, and a stack of locally hosted open-source models (Qwen 3 14B for voice-sensitive generation, Llama 3.2 3B for fast classification, Mistral Small for longer-form classification).
• Quality gates. Every drafted artifact passes five automated gates — voice match, factual support, hard-off-limits safety, puzzle solvability, brand presentation — before it reaches the operator queue. Items that clear every gate publish on schedule. Items that fail any gate are flagged for the operator with the specific failure.
• Operator review.The operator reviews anything the gates couldn't confidently clear. In launch month, every published artifact is human-reviewed regardless of gate status.
• Signal classification + clustering. Local models classify Stage 00 signals and cluster them into stories. Operator spot-checks the active set weekly and dismisses misfires.
• Case hero imagery — selectively.When a case ships with a sourced photograph (FLIR stills, archival scans, etc.), that's a real photograph with a real credit. When a case ships with a brand-plate or dossier-style hero instead, that's our work product over the case's own metadata, not a representation of a real document from the case. AI-illustrated atmospheric images, when we ship them, carry an ai_generated: trueflag in the case record and surface “AI illustration” in the caption. Every hero's sourcing is recorded in data/cases/SOURCES.md.

The list, also on the front of the page.

• Fabricated quotes. Every quote on the site traces to a real, cited source.
• Synthetic source documents. Every declassified file we reference is real, linked, and verifiable.
• AI imagery passed off as archival.When we use AI-generated illustration for atmosphere — rarely — it's labeled in the caption. Archival photos, FLIR captures, and document scans are real and credited.
• Auto-publish without gates. No content reaches /cases or /puzzle without passing the gate pipeline.

If we got it wrong, tell us.

Send corrections through the contact form with the case slug and the specific claim that's incorrect. Substantive corrections are made visible on the case page — the original text struck through, the updated text dated. Typo-level fixes happen silently. Cases corrected within 7 days of publication carry a small “corrected” chip in the byline row.

Skeptical of governments — not a vehicle for conspiracy bait.

Skepticism of specific government programs — including the framing or rollout of PURSUE itself — stays on the site. Some conspiracy literature is genuinely interesting to examine analytically (Operation Blue Beam in its non-antisemitic readings, the Bennewitz disinformation campaign, etc.); we engage with it as literature, not as fact.

What we will not publish, examine, or platform, even in “just asking questions” framing: anti-vaccine claims, “great replacement” theory, antisemitic-coded conspiracy framings (Rothschild / Soros tropes, antisemitic readings of Project Blue Beam), QAnon, or election denial. Submissions or partner content that touch these are declined without further discussion.

One person + a stack of tools.

Anomaly Daily is operated by one person. Press, embed inquiries, partnerships, sponsorships, corrections, and security disclosures all route through /contact with a structured form. No public email address by design — every inquiry gets a routing label so the right kind of message reaches the right place.