# Knowledge, research and evidence
canonical_url: https://www.3.14financialcontents.com/en/solutions/knowledge-research-and-evidence/
markdown_url: https://www.3.14financialcontents.com/en/solutions/knowledge-research-and-evidence/index.md
language: en
content_type: solutions
status: published
description: Establishing what is known, what sources support it, which versions are in force and what remains open.

Establishing what the organisation knows and what evidence can support it.

- **Ingestion and normalisation** — We gather heterogeneous sources and turn them into coherent structures.
- **Extraction** — We identify facts, entities, relationships, events, dates, figures and metadata.
- **Search and retrieval** — We combine lexical, semantic and structured search according to the nature of the information.
- **Connected knowledge** — We relate documents, concepts, people, organisations, decisions and events.
- **Assisted research** — We help to locate evidence, compare sources and keep visible the questions that are still open.
- **Provenance and review** — We preserve the relationship between a conclusion and the information used to build it.

## A knowledge system must go further

- identify the source
- preserve the context
- resolve entities and relationships
- recognise dates and versions
- compare information
- represent contradictions
- indicate when evidence is missing

## Result

The system can produce an answer, a structured record, a comparison, a piece of research, an alert or information reusable by other processes.

The system preserves provenance, contradictions and open questions so that a source can be examined, updated or challenged before its conclusions are reused.

## Qué podemos enseñar

- **Cómo funciona** — An ingestion layer normalises documents, tables and messages; extraction produces entities, facts and relationships while preserving the source position. Retrieval combines lexical search, vector search and structured filters, and every answer carries the fragments that support it. A version index keeps the revisions of a single document apart.
- **Qué podemos mostrar** — We can show a demonstrator over public corpora, the schema of entities and relationships, and the complete trace of a query: fragments retrieved, scores and the fragment finally cited. Also the documentation of the ingestion pipeline and of the duplicate-resolution rules.
- **Cómo se evalúa** — We measure recall and precision over sets of questions with known answers, and we review by hand whether the cited fragment really supports the claim. We track separately the abstention rate when the available evidence is insufficient.
- **Límites** — The system does not check whether the original source is correct: if the corpus is out of date, the answer will be too. Contradictions between sources are shown, not resolved. Entity resolution over uncommon or transliterated names still requires human review.
- **Estado** — operational

## Turning scattered information into knowledge that can be used and examined.

Information is usually found spread across documents, databases, messages, images, recordings and specialist knowledge.

The challenge lies in retrieving it and, beyond that, in recognising whether two sources are talking about the same thing, which version is in force, what relationship exists between them and what claims they can support.

Knowledge may end up in an answer, a piece of research or a structured record. It may also feed reporting, a decision, an agent or a process, provided it preserves the context and the evidence that new use requires.

## Entities, versions and the authority of sources

The same object may appear under several names. Two similar names may correspond to different entities. A rule may be valid only for one unit, one product, one period or one type of case.

Resolution does not consist of merging everything that looks alike. It consists of making explicit when two references can be treated as the same object and when they must remain separate.

Specialists establish which sources hold authority, define concepts and relationships, resolve domain ambiguities and decide between contradictory versions.

## Contradiction and abstention

Every claim needs evidence that supports exactly its meaning and its scope.

Contradiction is represented as knowledge. An answer can explain that two versions exist, show where they come from and point out which criterion is missing to resolve them.

Absence is represented too. When the available coverage does not support a conclusion, the system can abstain, formulate the pending question or propose which source would be needed.

## Graphs and logic

Graphs make it possible to navigate relationships that linear text hides.

```text
SOURCE → CLAIM → ENTITY → RULE → DECISION → OPERATION
```

Rules and deductions are kept explicit when they need reproducibility and review. DocuLogic is our line of work for reconstructing the logic built into software and turning it into examinable operational knowledge.

## Delivering knowledge to reporting, agents and decisions

Knowledge acquires more value when it can enter the place where a decision is taken or an action is performed. The same base can make it possible for:

- a person to receive the relevant context before deciding;
- an agent to retrieve valid tools, policies and evidence;
- a reporting system to reuse the same informational base;
- a predictive model to use consistent variables and definitions.

This solution applies informational intelligence to research and evidence: the common informational base, put to the service of establishing what is known and with what material it can be supported.

---

https://www.3.14financialcontents.com/llms-en.txt · https://www.3.14financialcontents.com/llms-full-en.txt