Is a FactBlock the same thing as a document chunk?

No. A chunk is an arbitrary slice of a document selected by similarity search. A FactBlock is a discrete factual proposition extracted from a source, with its own confidence score and validity window, independent of how the source document was split.

How is a FactBlock's confidence score calculated?

It reflects how well the extracted proposition is supported by its source text, and it's used across the API — for example the Fact Checker endpoint returns a confidence score for its verdict based on the FactBlocks and evidence it draws on.

What connects FactBlocks to each other?

The Temporal Knowledge Graph (TKG). FactBlocks are linked by typed causal relationships (causes, triggers, contributes to, prevents) and share a time-aware structure, so the graph can answer both what is true and when and why it became true.

Can I query FactBlocks directly?

Yes, through the Factagora API — endpoints like Fact Search, Fact Checker, and Evidence Finder all return results backed by FactBlocks rather than raw document text.

What Is a FactBlock?

The Smallest Unit of Trusted Knowledge

Most AI systems ground their answers in document chunks: arbitrary slices of text returned because they're semantically similar to a query. A FactBlock is different. It's Factagora's smallest unit of knowledge — a single, structured proposition extracted from unstructured text, carrying its own source, confidence score, and temporal context. Where a chunk is a fragment of a document, a FactBlock is a fact that can stand on its own and be checked.

From Unstructured Text to FactBlock

News articles, filings, and research are parsed into individual factual claims, each attached to its own metadata rather than inheriting the metadata of the document it came from:

●Proposition — the fact itself, stated as a discrete claim (e.g. "Company X raised a $50M Series B")
●Source — the origin document or article the fact was extracted from
●Confidence — a score reflecting how well the extraction is supported by the source
●Temporal context — when the event occurred and the window during which the fact stayed valid

Why FactBlocks Connect Instead of Sitting Alone

FactBlocks aren't stored as an unordered pile of facts. They're linked in a Temporal Knowledge Graph (TKG) by semantic relation and time. Each fact is bi-temporal — it records both when the event occurred and the window during which it stayed valid — so queries return only what was true at the moment being asked about. FactBlocks are also connected by typed causal links (causes, triggers, contributes to, prevents), which are promoted from hypothesis to verified as evidence accumulates.

Unstructured text is decomposed into FactBlocks, each carrying its own source and timestamp, then linked into a Temporal Knowledge Graph.

FactBlock vs. a Vector Chunk

●A vector chunk is retrieved because it's similar to the query. A FactBlock is retrieved because it's the fact that matches the query.
●A chunk has no notion of time. A FactBlock is bi-temporal — valid from one point to another.
●A chunk carries no verdict. A FactBlock carries a confidence score and can be checked against other FactBlocks that support or contradict it.
●For more on why that gap matters, see RAG Doesn't Cure Hallucinations.

Where FactBlocks Show Up in Factagora

Fact Search returns FactBlock-backed news, not raw articles. Fact Checker verdicts are backed by related FactBlocks — the `sources` field in its response points to actual claims and predictions, each with its own credibility score. Evidence Finder surfaces FactBlocks tagged by stance (supports, contradicts, or neutral) instead of just topically related text. See the API introduction and Search API overview for the full endpoint list.