About
How We Attribute Sources
Every narrative on AIDRAN is generated from real public discourse — posts, comments, papers, and articles collected from seven platforms. None of it is invented. But not every claim in a narrative carries a citation, and that is by design. This page explains why, and what the markers you see in our stories actually mean.
Two Kinds of Claims
The core distinction in our attribution system is between claims drawn from the collective voice of many sources and claims tied to a specific individual or organization. Each is handled differently.
Cohort Voice
No citation neededWhen three or more independent sources express the same sentiment on the same topic, we synthesize their collective position without citing any single one. The entire body of discourse backs the claim.
- —"Researchers were skeptical of the timeline."
- —"The AI safety community pushed back hard."
- —"Reaction across Bluesky was mixed, with enthusiasm in some corners and alarm in others."
Named-Source Attribution
Citation requiredWhen a narrative assigns a specific position, claim, or statement to a named person, organization, or publication, that claim must trace to an actual discourse record in our database.
- —"Yann LeCun argued that current architectures cannot achieve general intelligence." [1]
- —"A Nature paper concluded that scaling laws plateau beyond a certain parameter count." [1]
- —"The company said in a March 14 blog post that the feature would ship in Q2." [1]
The deciding question is always: are we describing something visible across many voices, or assigning a position to a specific entity? If the former, no citation. If the latter, a numbered marker appears.
The Fidelity Score
Every narrative receives a fidelity score between 0 and 100. This is not a truth rating — it measures how well the narrative’s specific claims are anchored to identifiable source records.
A high fidelity score means the narrative’s named attributions can be traced to specific discourse records with verbatim supporting passages. A lower score means some claims rely on broader synthesis or could not be linked to a single record in the source pool.
What the score reflects
- •Sourcing specificity — can each named claim be traced to a record?
- •Attributable claim density — what proportion of claims carry verifiable sources?
- •Epistemic transparency — does the narrative signal its own uncertainty honestly?
The Evidence Badge
This badge appears on every story. Click each element to see what it communicates.
Click any part of the badge to learn what it means.
Attribution Tiers
Sources attached to a narrative fall into one of three confidence tiers. These are displayed as colored indicators alongside the source list on story pages.
Cited
The claim traces directly to a specific discourse record in the source pool. A verbatim supporting passage is on file. This is the highest confidence tier.
Inferred
The source was reconstructed after the fact through embedding similarity — the record most likely to have been in the original generation context, matched by vector proximity and timestamp. Reliable above 0.85 cosine similarity.
Synthesized
The claim is drawn from the collective discourse pool rather than any single record. It represents a pattern visible across multiple sources. No individual citation applies.
Older narratives published before our provenance tracking system went live may show more inferred sources. A note appears on those stories explaining that direct citations are available in newer coverage.
Fidelity Flags
Occasionally you will see a [?] marker after a claim instead of a numbered citation. This is a fidelity flag.
A fidelity flag means the narrative included a specific attribution that could not be traced to a discourse record in the current source pool. The system chose to keep the claim because it was important to the story’s context, but it flagged the gap honestly rather than fabricating a source or silently dropping the attribution.
What fidelity flags are not
- •They do not mean the claim is false.
- •They do not mean the narrative is unreliable.
- •They do not indicate an error in the system.
A flag means the claim drew on knowledge beyond the immediate discourse pool — often well-known public facts or widely reported statements. It is a signal that independent verification is warranted, not a judgment on accuracy.
Attribution by Narrative Type
Different story formats serve different editorial purposes, and attribution density varies accordingly.
Lead Stories
Lead stories report on significant events and statements. They carry the highest density of named-source citations because they make the boldest, most specific claims.
Beat Narratives
Beat narratives synthesize ongoing discourse around a topic over time. They blend cohort voice with anchored citations for key moments, producing a natural mix of both registers.
Dispatches
Short-form dispatches favor cohort voice for brevity. Named citations appear only when a specific attribution is genuinely striking or essential to the dispatch.
Entity Narratives
Entity profiles are dense with named attribution by nature — everything relates to what a specific person or organization said, built, or argued. Citation markers are critical here.
The Short Version
When many voices say the same thing, we write it as collective sentiment and cite no one — because citing one would misrepresent the breadth of the signal. When a specific person or organization is named, we trace the claim to a real record and mark it with a numbered citation.
When we cannot trace a specific claim but believe it belongs in the narrative, we flag it honestly. A flag does not mean the claim is wrong. It means independent verification is warranted.
Transparency about what we know, what we inferred, and what we could not verify is the foundation of every narrative we publish.