How We Attribute Sources

AIDRAN writes stories from public source material stored in its corpus and, when enrichment adds them, public web links. The current system has two related source surfaces: the corpus records that fed story generation, and the citation rows selected during enrichment. Delivery expands those into the numbered markers, inline links, excerpts, and source drawers you see on story pages.

Current stories use two citation modes, depending on where the source lives.

The visible rule is simple: internal source records get bracketed markers; external web sources get inline links.

The core distinction is between a claim that belongs to a specific source and a claim AIDRAN is making from the pattern across many sources.

• Citation required: direct quotes, exact dates, statistics, official announcements, attributed statements, and factual claims assigned to a named person, company, publication, paper, or regulator.
• No marker required: AIDRAN’s own editorial analysis, structural conclusions, and broad descriptions of a pattern visible across the story’s source set.
• Use precise attribution: if a source is a comment, a blog post, or a forum thread, the story should not inflate it into an institutional research claim.

That is why a story can have both cited facts and uncited interpretation. The citation marks the evidence for a source-level fact; it does not outsource the story’s thesis to that source.

The public story page fetches source details separately from the story body. When source data is available, the floating Sources button opens a drawer that combines two buckets:

• Corpus records expanded from the story’s stored source record ids. These rows show the source kind, title or URL, byline when captured, publish time, sentiment score when available, and a short excerpt from the record body.
• External citations from enrichment citation rows that contain a web URL instead of a corpus record id. These rows show the domain, title or excerpt, and, when emitted, the specific story claim the link supports.

Drawer numbering is a reading aid. Corpus records appear first, followed by external web citations. A visible [N] marker in the story body is an internal citation marker; a drawer row is the inspectable support record or citation link currently returned by Delivery. Some rows may represent broader generation context rather than a one-to-one footnote for a visible marker.

For corpus records, excerpts are short previews from the stored record text. For external web citations, excerpts and claim labels are enrichment metadata. Neither replaces opening the original URL when you need the full source.

The backend keeps three ideas separate:

• Generation context lives on the story as source record ids: the corpus records that fed the story’s generation and source-volume displays.
• Citation rows live in the citation table. They carry marker numbers for internal citations, external URLs for web citations, excerpts, claim labels, confidence fields, and source metadata when enrichment emits them.
• Signal provenance can appear as story metadata explaining why AIDRAN flagged the story in the first place: signal kind, severity, score, timestamp, and detector confidence. It is not a source citation.

The current public schema and story UI are intentionally narrower than a full claim audit system. The page does not show:

• A claim-by-claim truth score
• A public three-tier source rating
• A guarantee that every sentence maps to one drawer row

The system does retain per-citation metadata such as inferred and confidence when Editorial emits it. Those fields describe individual citation rows; they are not the same thing as a public source-rating model.

Different story formats serve different editorial purposes, and attribution density varies accordingly:

• Lead stories usually carry the densest visible sourcing because they make broader claims about significant events.
• Secondary stories tend to anchor the immediate signal with a smaller set of concrete sources.
• Beat stories blend pattern language with anchored citations for key moments inside an ongoing topic.
• Dispatches are brief, but concrete dates, official announcements, direct quotes, and named factual claims still need citations.
• Entity stories often carry more named attribution because the format focuses on a specific person, organization, product, or concept.

Numbered markers point to internal corpus-record citations. Inline links point to external web citations. The source drawer shows the support set Delivery can currently expand for the story: corpus records first, external web citations after them.

A citation is required when a story makes a specific factual or named-source claim. AIDRAN’s own analysis and broad pattern language may be grounded by the full source set without attaching every sentence to a single source.