Anomaly Index

Methodology

Anomaly Index is government-source-first. This page sets out how an incident qualifies for inclusion, how sources are weighted, how a classification is decided, and how the record is corrected when it is wrong.

Coverage

Anomaly Index is a curated global selection of documented incidents from 1940 onward. It does not claim to contain every historical report. Anomaly Index currently indexes qualifying incidents found within the official collections listed here.

That distinction matters. This is not an inventory of every UFO or UAP report ever filed, and it is not a random sample of them. It is the set of incidents that (a) fall within one of the source collections described on the Collections page, and (b) meet the inclusion criteria below. As more official collections are processed, the map will grow. As of this writing it indexes 25 incidents drawn from 52 cited sources; that is a snapshot, not a ceiling, and it should not be read as a claim about how frequently anomalous events occur.

Source policy

Every substantive claim on Anomaly Index is attributed to a source record, and every source record carries a tier. The tier determines what a source is allowed to do editorially.

Tier A — primary sources

Government records, official investigation reports, declassified documents, original photographs or recordings, official hearing transcripts, government archival collections, and academic papers containing original analysis or data.

Tier A is the only tier that can qualify an incident for the main map. At least one accessible Tier A source is required before a case is created.

Tier B — credible secondary sources

Established investigative reporting, major wire services, reputable newspapers and broadcasters, and scholarly historical work.

Tier B sources may clarify context, identify disputes between other sources, connect records to one another, or help locate primary material. They should rarely determine a high-confidence editorial classification by themselves, and they can never substitute for the Tier A source an incident needs to qualify in the first place.

Tier C — discovery and context only

Documentaries, advocacy organizations, UFO research organizations, books without accessible sourcing, interviews, personal websites, and social media posts.

Tier C material can point researchers toward primary records that haven't yet been located, and it can supply useful context once a case already qualifies. It cannot independently qualify an incident for the main map, no matter how detailed or well known it is.

Inclusion criteria

An incident is added to the map only when all of the following hold:

Deduplication is treated as an editorial task, not an afterthought. Multiple records describing the same event — a witness statement, a radar log, and a later investigation, for instance — are folded into one incident so that repeated documentation of a single event does not inflate the map count.

Location precision

Every incident states its location precision explicitly, using one of:

exact
approximate
city
regional
country_only
withheld

Sensitive-infrastructure policy

An incident is tagged as associated with a nuclear or military site only when an official source explicitly identifies the facility or operational context — never from geographic proximity alone.

For incidents near active sensitive sites, the record:

Classification system

Every incident carries two separate fields, and the site never blends them into one:

official_disposition
index_assessment

official_disposition is what the investigating body actually concluded, in terms as close to their own as possible — including "unresolved," "no published determination," or a specific identification. index_assessment is Anomaly Index's own editorial classification, always labeled as editorial and never presented as an official finding. For example:

Official disposition: No published determination
Anomaly Index assessment: Likely identified

An official disposition and an editorial assessment are frequently different things, and both are shown. Every revision to an editorial assessment is recorded in that incident's classification history rather than overwritten.

The four levels

  1. Identified — a specific conventional explanation is supported by strong evidence (balloon, aircraft, bird, satellite, astronomical object, sensor artifact, optical effect, and similar).
  2. Likely identified — one conventional explanation fits the available evidence better than the alternatives, but direct confirmation is missing.
  3. Unresolved: insufficient data — the evidence is too incomplete, degraded, or contradictory to support a reliable identification.
  4. Unresolved: corroborated — multiple credible, partly independent evidence channels document the event, but evaluated explanations do not adequately resolve it. This category has a deliberately high threshold; it is not a general "interesting case" bucket.

Of the 25 incidents currently indexed, 4 are classified identified, 5 likely identified, 11 unresolved (insufficient data), and 5 unresolved (corroborated). Those counts will shift as records are added or reviewed — they describe the current state of the repository, not a fixed target.

Assessment decision tree

Every editorial assessment is reached by working through the same four questions, in order:

  1. Does strong evidence directly support a specific conventional identification? → Identified
  2. Does one conventional explanation account for most of the evidence without direct confirmation? → Likely identified
  3. Is the evidence too incomplete, degraded, or contradictory to distinguish between explanations? → Unresolved: insufficient data
  4. Do multiple credible, partly independent evidence channels document an event that evaluated explanations fail to resolve? → Unresolved: corroborated

The word unresolved is never used as a synonym for extraordinary, extraterrestrial, or nonhuman. Every unresolved case carries this statement or a contextually equivalent version of it:

Unresolved means that the available evidence does not support a reliable identification. It does not imply extraterrestrial or nonhuman origin.

Corrections

Anomaly Index treats corrections as a normal, expected part of maintaining a public record, not an exception. If something is wrong — a broken source link, a misdated event, a classification that no longer fits the evidence — it should be reported and fixed.

When the repository opens to public contributions, every case dossier will include:

During the private repository release, the privacy page states the corrections policy and explains that public intake is not yet open. Corrections backed by a primary source are the most valuable contribution this project can receive. Classification changes are appended to a case's classification history rather than silently overwritten, so the record of how an assessment changed over time is preserved along with the current assessment itself. All changes, whether from the maintainer or a contributor, go through editorial review before they are published — repository access does not mean community votes determine factual status.

The repository is the canonical data surface

Anomaly Index does not have a separate data-download page or export button. The private GitHub repository is the canonical data store during the current release, and the public website is its rendered view.

Path Contents data/incidents/ One YAML record per incident — the full evidence record: what was reported, what's verified, the evidence ledger, proposed explanations, classification history data/sources/ Cited source records, each with a tier (A/B/C), organization, and, where applicable, a collection ID data/stories/ Guided-story narratives and their map camera state data/facilities/ Facility context records, limited to already-public information schemas/ The JSON Schemas every record must validate against scripts/ Validation, deduplication, GeoJSON generation, search-index generation, and link-checking

At build time, every record is validated against its schema, source references are checked, dates and coordinates are checked, probable duplicates are flagged, and external links are checked against archived alternatives. The build fails on structural errors; editorial warnings are reported without blocking a local preview. Anyone can clone the repository, run the same validation, and see exactly how a number on the map was produced.