Relationship graphs are persuasive. A picture of a company, its owners and its directors suggests that the links are facts. In due diligence they are claims, made by a registry on a date, by a filing, by a document in the data room or by management. A graph that forgets where a link came from turns a claim into an apparent fact.
The statement is the unit of truth
AnKre stores the knowledge graph as a log of statements. Each statement records what is asserted, which source asserted it, when it was retrieved, the period it refers to and what produced it, whether a collection step, an agent or a person. The nodes and edges you see are views computed from those statements; nothing is written to the picture without a statement behind it.
Three dates are kept apart because they answer different questions: the date of the fact, the date of the document and the date of collection. A shareholding valid in 2023, recorded in a filing of 2024 and retrieved in October 2026 is three facts, not one.
History is never overwritten
When a source changes, or a reviewer corrects an assertion, AnKre adds a statement; it does not erase the previous one. This is what allows a file to be replayed as it was known at a given date, and what lets a committee see why a conclusion was reasonable when it was reached, even if a later record changed the picture.
Contradictory versions are shown side by side, never silently merged: this source says one thing, that one another, and the difference is stated until someone settles it.
Identity resolution with a human in the loop
The same company or person appears differently across registries, filings and documents. AnKre resolves those appearances into one subject using strong identifiers first, such as registration numbers. Uncertain matches go to a person for review, and every merge can be undone, so a wrong identity does not quietly contaminate the findings that depend on it. When an identity is corrected, the conclusions it affects are shown.
Your file and public reference data stay apart
Material gathered for a client's file belongs to that file. Public reference data about legal entities is shared, read-only, so that every file benefits from it without any file's content leaking into it. The two are joined when a question needs both, never stored as one graph.
The consequence for a reader is simple: any line on the graph can be opened to the statements behind it, with their sources and dates, and a link that is hidden by a filter is never presented as a link that does not exist.