Data Methodology

The Legacy Archive NUFORC dataset originates from the National UFO Reporting Center, founded by Robert Gribble in 1974 and directed by Peter Davenport since 1994. NUFORC publishes its full case database as a downloadable CSV export. The version used for this import contains 141,271 individual report records spanning from the early 1900s through the present, including historically reconstructed cases submitted by researchers and original-source witness reports filed directly with NUFORC. Note: the raw source file contains 928,663 physical lines, but many witness narratives span multiple lines within a single quoted CSV field; the true logical record count, verified by parsing the file end-to-end, is 141,271.

Each row in the source CSV includes: a dossier identifier, sighting summary, full witness narrative, shape (free text), and a single concatenated stats field containing structured metadata in plain-text form.

Known source data quality issues — the raw CSV is not free of noise: MADAR sensor network node logs are mixed in with witness reports (identifiable by titles like "MADAR Node 100"); some records contain NUFORC editorial annotations such as "((NUFORC Note: Source provides no detailed information))" or "((HOAX??))"; non-English location text (notably French GEIPAN-overlap cases) contains character encoding artifacts from the source export; and the free-text shape field uses 25+ overlapping terms (e.g. circle, disk, orb, sphere all describing similar morphology) reflecting decades of inconsistent data entry by different NUFORC staff and eras.

NUFORC's public CSV export has been available for years, and individual reports have always been readable. What had not been built — by NUFORC, by any third-party mirror, or by any prior research tool — was a system capable of parsing the stats field into queryable structured data.

That field follows a consistent internal format across the entire dataset:

This single text block contains three separate date/time events (when the sighting occurred, when it was reported to NUFORC, when NUFORC published it), a duration value, and — critically — a comma-separated list of 15 standardized characteristic phrases that NUFORC has applied with remarkable internal consistency across its entire multi-decade dataset. To the best of our knowledge, no prior public tool had exposed these as queryable fields before this project.

Date extraction — three separate datetime fields are parsed via regular expression from the stats block: Occurred date/time (the event itself), Reported date (when the witness submitted to NUFORC), and Posted date (when NUFORC published the case). Dates are normalized to ISO 8601 format on import. Where only a date is present without a time, the time defaults to midnight and is flagged accordingly in the record.

The 15 NUFORC-native characteristics — these are extracted via exact substring matching against the literal phrases NUFORC itself uses in the stats field. No interpretation or inference is applied; a characteristic is marked true if and only if its exact NUFORC phrase appears in that record's stats text. The 15 phrases:

The 15 characteristics listed above are extracted directly from NUFORC's own classification system. The two additional fields described below (Military Proximity and USO) were created by AlienAlerts.com and are not part of NUFORC's original data structure.

Two derived (non-native) characteristics — Military Proximity and USO (Submerged/Maritime) are not present in NUFORC's own data structure. These are classifications independently derived by AlienAlerts.com through keyword search against the full witness narrative text. This distinction is important for forensic purposes: the 15 characteristics above are NUFORC's own categorization of the report; these two are our own pattern-matching layer applied on top of the source text, and carry a higher risk of false positive/negative than the NUFORC-native fields.

Example false positives — a report mentioning "the ocean view from my hotel window" will be flagged USO-positive despite containing no maritime sighting content. A report describing a witness who "used to work military" decades before their sighting will be flagged Military Proximity-positive despite no actual base or installation involvement in the event itself. Both derived fields use simple case-insensitive substring matching with no contextual disambiguation, and should be treated as a coarse first-pass filter requiring manual verification — unlike the 15 NUFORC-native fields, which reflect NUFORC's own original classification of the report.

Live statistics computed against the currently imported subset of the dataset (import is ongoing as of this writing):

Records lacking coordinates (~20%) are typically those where the witness-supplied location text could not be geocoded by NUFORC's own export process — vague descriptions, non-standard place names, or international locations without sufficient detail. These records remain fully searchable by text and date but will not appear on the map interfaces.

For researchers establishing the boundaries of what can and cannot be derived from this dataset:

• No witness contact information or personally identifiable data of any kind
• No photographs, videos, or physical evidence attached to records
• No follow-up investigation notes from NUFORC beyond what already appears in the original published narrative
• No formal classification of report credibility, reliability, or hoax probability — except occasional NUFORC editorial annotations that are preserved verbatim within the original source text where present
• No corroborating radar, military, or third-party confirmation data — this is exclusively civilian self-reported witness data as collected by NUFORC

This parsed dataset and the accompanying search, mapping, and analysis tooling are intended for researchers, investigators, and analysts conducting pattern analysis and historical review across a large body of civilian UAP reports. It is not intended as a substitute for original source verification on any individual case, and should not be cited as confirmation of any specific claim without independent review of the underlying NUFORC report.

Every imported record retains a link to its original NUFORC source case page. Researchers can cross-reference any AlienAlerts.com record against the original NUFORC entry to verify parsing accuracy. No witness narrative text is altered, summarized, or paraphrased during import — the full original text is preserved verbatim. Only the structured metadata (dates, characteristics, shape) is extracted into separate searchable fields; the source text itself remains unmodified and viewable on every record page.

The parsing logic is consistent and deterministic — identical input text will always produce identical extracted output. No machine learning, fuzzy matching, or probabilistic classification is used for the 15 NUFORC-native characteristics; they are exact substring matches against NUFORC's own stated text. This was a deliberate design choice to keep the primary dataset forensically defensible.