Every item the AI learns is classified T1–T4 by source reliability — a judgement about where the evidence came from and how authoritative that origin is. This tier is kept deliberately separate from an individual item's quality score, which measures the condition and clarity of that specific item. Tier tells you how much to trust the source; quality tells you how usable the artifact is.
Live tier distribution across everything the AI has learned. Bar length reflects each tier's relative share of the corpus.
How the AI classifies a source, the kinds of origins that qualify, and how each tier is weighted and labelled in an answer.
Direct government and primary-source releases — the most authoritative origin in the corpus. Material that comes straight from the issuing body, without an intermediary retelling it.
NARA RG 615 declassified files, AARO reports, DVIDS military releases, ODNI assessments, congressional records and hearing transcripts.
Highest confidence. Weighted most heavily and labelled T1 in answers. Treated as authoritative on what was officially stated or released.
Mainstream and institutional reporting, plus named first-hand records. Not the issuing body itself, but a credible, accountable intermediary with editorial or institutional standards.
Established news organisations, peer-reviewed and institutional publications, official incident write-ups, and named, on-record first-hand accounts.
High confidence. Weighted strongly and labelled T2. Trusted for reporting, but distinguished from the primary release itself.
Researcher and community-vetted material. Serious independent analysis that has been reviewed or cross-checked by a knowledgeable community, but lacks official or institutional backing.
Independent researcher reports, community-verified case reconstructions, curated datasets and analyses vetted by established research groups.
Moderate confidence. Weighted with care and labelled T3. Useful for corroboration and context, flagged as unofficial.
Anecdotal, uncorroborated, or opinion-based material. The origin cannot be independently confirmed, or the item is inherently subjective.
Opinion pieces, forum posts, uncorroborated witness or whistleblower claims, and user uploads. These are hard-capped at T4 until corroborated by a higher tier.
Lowest confidence. Weighted least and labelled T4. Surfaced as unverified — never presented as established fact.