AI can suggest a useful structure for a passage: a person, a company, a role connecting them and an excerpt that explains why. The important moment comes next, when you decide what deserves a place in your Network.
Lage Bureau's integrated AI assistance follows that sequence. You supply text, inspect a proposal and explicitly apply the items you choose. Generation is an input to the investigation. Your assessment of the source and the connection remains yours.
This walkthrough uses a fictional company notice and an intentionally unsupported ownership interpretation. It contains no real company finding or personal information. The example is small enough to compare each proposed relationship with the words that prompted it.
Choose the passage and the context
Start from the company case's AI checkpoint and paste the complete English officer notice. It includes these statements:
LB Training — Mara Voss (training person LB-P-001) is a director of LB Training — Merebridge Ltd.
LB Training — Owen Reed (training person LB-P-003) is also a director of LB Training — Merebridge Ltd.
The notice explicitly identifies Mara as the same authored person, LB-P-001, in the German Lindenwerk notice. It makes no ownership or beneficial-control assertion. The task is to add the UK officer details without promoting the shared-director clue into ownership.
AI extraction uses a tester-owned OpenRouter API key. Use a separate spend-limited key for the prototype. The browser sends your pasted text and any Network context you explicitly select to the configured provider workflow. The key is stored unencrypted in the app's browser settings, so use synthetic or non-sensitive text and keep credentials out of recordings and shared files.
Select the existing Merebridge and Mara records as context, as the checkpoint's review guide specifies. Owen is the new person to look for. Check whether every reference points to the intended entity. The explicit exercise identifier supports Mara's match; a similar label alone would not.
Read the proposal as a set of decisions
The result may propose entities, relationships, Sources and Citations. Look for Mara-to-Merebridge and Owen-to-Merebridge director relationships with exact supporting excerpts. Output varies by model and request; the authored review guide is not a captured response or a guarantee of those results.
For each proposed relationship, inspect three things together:
- The endpoints. Are the person and company the records you intended?
- The relationship type. Does “is director of” reflect the passage, or has the model proposed a stronger claim?
- The excerpt. Do the quoted words actually support that relationship?
An ownership proposal should not be accepted simply because Mara connects Lindenwerk and Merebridge. The supplied notices leave ownership undetermined. Reject that interpretation if it appears, while reviewing the two UK director relationships on their own merits.
This is why a reviewable proposal is more useful than an unexplained finished picture. You can inspect the suggested structure before it becomes part of your working document.
Understand what validation establishes
Lage Bureau checks the proposal's structure and references before it is ready to apply. Those checks include valid kinds and endpoints, dependencies and the exact excerpts anchored in the supplied text. A relationship proposal needs supporting text that can be found at the recorded position.
These checks answer a bounded question: is this a well-formed proposal with the required local support? They do not establish that the original notice is authentic, current or accurate. They also do not prove that a model's interpretation of a valid excerpt is the best one.
| Check | What still needs your judgment |
|---|---|
| An excerpt matches the pasted text | Whether the source itself is trustworthy and relevant |
| A relationship has valid endpoints | Whether those endpoints represent the intended people and companies |
| The proposed type is supported by the registry | Whether that type expresses what the passage actually says |
| A model supplies a confidence value | Your own confidence in the durable record |
Machine confidence is proposal metadata. It does not automatically become analyst confidence. Accepted entities and relationships default to unconfirmed unless you deliberately select a different assessment.
Review connected items together
A relationship needs its endpoint entities. A Citation needs a Source and a target. Those dependencies matter when you accept only part of a proposal.
If you reject a newly proposed person, a relationship that depends on that person cannot be applied as though the person had been accepted. Review the explanation for blocked items rather than trying to preserve a connection with a missing endpoint. Likewise, keep the supporting Source and Citation with the relationship you intend to add.
Choose the dependency-valid set you want, then apply it. The accepted batch becomes one user-triggered, undoable change. Undo lets you reverse that application; it does not reverse the provider request or any cost already incurred by generation.
Inspect the result in the ordinary editor
After applying, select a relationship in the Network. Check the label, confidence and attached citation again. The proposal should have become an inspectable research record, with its source still available to explain it.
Now test an alternative interpretation. Add an analyst note about what the passage does not establish, or leave an uncertain relationship unconfirmed while seeking better material. Keep a proposed identity association separate when the evidence is ambiguous. The fictional identity case offers a second exercise in resisting a stronger conclusion than the sources support.
The integrated AI workflow does not browse the web, retrieve a registry, upload a PDF, merge identities or write to your Network without acceptance. Its useful role here is to propose structure from text you supplied and make that structure available for your review.
Open the cross-border company case to work through the same passage, or read how Lage Bureau works. Keep the question small enough that you can explain every accepted connection.
About the source
The quoted company notice is authored training text. It supports this exercise in reviewing a proposal, not a conclusion about any real officer or company. Read the passage beside the suggested relationship; that comparison is the centre of the exercise.