Encounters are not pages
An encounter is the claimed real-world event. A source record is a report, article or episode that describes it. One event can be told in several places, and one report can describe several events — a witness often mentions an earlier experience, or one a relative had. The Anomadex keeps these apart: each event becomes its own entry, labelled as the main report or a related account, and linked back to the report that describes it.
Source records are stored as versions and never edited in place. When a source changes a page, a new version is stored and the old one is kept.
Every fact has a source
Each value on an entry page names the report it came from and the part of that report — a labelled field, a named section, the headline. It also says how the value was obtained:
| Label | Meaning |
|---|---|
| Stated | Stated in the source record. |
| Derived | Worked out from stated values by a documented rule. |
| Checked by hand | Entered or corrected by me after checking the source. |
| Geocoded | Placed on the map from a place name. |
| Inferred | Read from the narrative by the extraction model; lower certainty. |
When two statements disagree — a report's labelled field says one height and the narrative another — both are shown. One is chosen for display, and the other appears beside it as a conflicting value. Nothing is silently overwritten.
Dates and places, only as precise as given
A date is shown at the precision the source gives: a day, a month, a season, a year, or a phrase such as “two years before” for events described relative to another. The Anomadex never invents a day to fill a gap.
Places follow the same rule. On the map, a solid dot is a point the source itself gives (coordinates in the report, or the source's own map). A ring is an estimate from the nearest town or the centre of a county, and every point carries its uncertainty radius.
| Map point | How it was placed |
|---|---|
| Coordinates given in the report | SOURCE_PROVIDED |
| The source's own map point | SOURCE_MAP_LAYER |
| Same place as a related event | SAME_AS_RELATED |
| Named feature, geocoded | GEOCODED_NAMED_FEATURE |
| Nearest town, geocoded | GEOCODED_LOCALITY |
| Centre of the county | COUNTY_CENTROID |
| Placed by hand | MANUAL |
Ratings and quality
There is no single “credibility score”. Separate questions get separate answers:
- Source classification is copied exactly as the source publishes it. BFRO, for example, rates reports Class A (a clear sighting where misidentification is judged unlikely), Class B (something heard or seen less clearly) and Class C (second-hand or less reliable). These are the source's ratings, not the Anomadex's.
- Account type records how close the account is to the witness — first-hand, second-hand, an investigator's summary, folklore.
- Extraction confidence measures only whether the source was read correctly, not whether the claim is true.
| Account type | Meaning |
|---|---|
| First-hand | The witness describes what they personally experienced. |
| Direct interview | An interviewer recorded the witness directly. |
| Second-hand | Told by someone who heard it from a witness. |
| Third-hand | Passed through two or more people. |
| Investigator summary | A researcher summarizes a witness report. |
| Press report | A journalistic account. |
| Folklore / legend | A traditional account, not a modern witness claim. |
| Unknown | The source relationship could not be determined. |
Machine reading, checked by code
Labelled fields — report number, date, county, town, classification — are read by ordinary parsing code. A language model reads only the free-text narrative, to identify what kinds of observation were described, the creature description, what the witness was doing, and to draft the project-written summary.
Its output is checked: every quoted piece of evidence must actually appear in the report, or the value is dropped and sent to a human reviewer. Values from labelled fields always win over values the model proposes. Names of submitters are removed before anything is stored.
Privacy and sensitive locations
- Witnesses are never named. Submitter names, phone numbers, email addresses and similar details are not stored.
- Where a location is sensitive — typed by a witness, or on private land — the public point is rounded to about 1 km. Some locations are withheld entirely.
- The Anomadex does not try to identify anonymous witnesses by combining sources.
- If you accept analytics cookies, visits are measured with Google Analytics, which counts pages viewed and searches made; nothing is loaded or set until you accept, and declining changes nothing else about the site. It is not linked to anything you send through the correction or contact forms. Change your choice any time with .
Sources and attribution
The current catalogue is built from the Bigfoot Field Researchers Organization (BFRO) Geographical Database and the Sasquatch Chronicles podcast. The Anomadex stores structured facts, a link and a fingerprint of each report or episode — not the witness's narrative or the episode transcript. To read or hear an account in the witness's own words, follow the link to the original report, or the episode link, which starts playback where the account begins.
Sources are collected politely: at most one request every few seconds, and only public pages.
Corrections and takedowns
Every entry page has a Suggest a correction form. Reports go to a review queue tied to that entry and its source record, and I check each one against the source. Use the same form for privacy or takedown requests.
Contact
For anything the correction form doesn't cover — a privacy or takedown request you'd rather not send through the form, a question from a source publisher about how its material is used, or permission and licensing matters — use the contact form. If your message is about a particular entry, include its web address so I can find it quickly.