How UFOIntel Treats Evidence

• A statement is treated as evidence only when it can be tied back to a primary or archival document that exists in the corpus. Everything else is clearly labeled as discussion, context, or pending material.
• The platform leads with the source. Where the document is, who released it, and what it literally says comes before any interpretation layered on top of it.
• When the corpus has no supporting document, the platform shows an honest empty or pending state rather than filling the gap with a plausible-sounding claim.

• No invented documents or evidence. The system will not synthesize a document, quote, case file, or sighting that is not actually present in the corpus.
• No invented map pins. A location is never placed on the map unless it derives from a real, document-backed extraction that passes the credibility rules below.
• No invented sources, users, posts, or metrics. Counts and statistics are read live. When there is no data, the value is reported as null or labeled “Pending — no corpus ingested yet” / “Not configured” — it is never rounded up from zero.
• Service health (database, search, storage) is reported from real configuration and connectivity. The platform does not claim a service is healthy when it is not.

Sources are tiered by their evidentiary strength. A declassified government release, a FOIA-produced record, and an uncorroborated forum post are not interchangeable, and the platform does not present them as such.

• Credibility reflects the provenance and verifiability of the source — official archives and primary releases rank above secondary aggregation, which ranks above anonymous claims.
• Confidence reflects how directly a given document supports a specific claim, including whether the supporting text was machine-extracted (e.g. via OCR) and how clean that extraction is.
• Scores are surfaced to readers (for example via score meters) so the strength of the underlying evidence is visible, not buried. Low confidence is shown as low confidence.

Traceability is the core contract of the platform. A public, evidence-bearing fact is expected to carry the full chain back to its origin:

If any link in that chain is missing, the item is not presented as verified evidence.

Geographic pins are held to a stricter bar than text, because a pin on a map reads as a confident factual claim about a place. Every candidate pin is run through classifyPinCredibility, which assigns one of four tiers:

• Coordinate-range guard: latitude/longitude must fall within valid Earth bounds. Out-of-range coordinates classify a pin as invalid.
• OCR-evidence guard: a pin must trace to extracted document text (the supporting snippet). A coordinate with no document evidence behind it cannot become credible.

Ingestion is the path a document takes from a source archive into the searchable corpus. The pipeline is designed as follows, and is gated / not yet live — no documents have been ingested.

1. 1Queued. Acquired documents are enqueued for processing rather than indexed on arrival.
2. 2OCR. Scanned and image-based documents have their text extracted so claims can cite literal snippets.
3. 3Deduplication.Documents are dedup'd by content hash and a dedupeKey so the same record is not counted or shown twice.
4. 4Embeddings. Text is embedded to support semantic similarity and relationship discovery.
5. 5Search index. Processed documents are written to the search index so they become discoverable.

Until this pipeline is enabled, corpus-backed views are intentionally empty and labeled as pending.

• A successful upload is recorded as received / pending processing. The platform never claims an uploaded file has been processed, verified, or admitted to the corpus on submission.
• Uploaded material does not become evidence until it has passed review and the same ingestion and traceability rules as any other source.
• Until then it is not shown as a verified document, is not counted in corpus metrics, and cannot back a public fact or a map pin.

• Community is discussion, not corpus evidence. Posts, comments, and claims made by users are kept structurally separate from verified, document-backed records.
• A community claim does not gain evidentiary status by being upvoted, repeated, or popular. It only counts as evidence if and when it is backed by a traceable document in the corpus.
• Community features are currently a preview — posts are not yet persisted. The platform does not render fake user activity to make the space look populated.