A company record can arrive as a short register-style entry, a spreadsheet row or a paragraph in a notice. The research question survives the change of format: which company does this describe, what does it tell you, and which connections can you support?
This walkthrough uses three fictional companies: LB Training — Lindenwerk GmbH in Germany, LB Training — Atelier Rive SAS in France and LB Training — Merebridge Ltd in the United Kingdom. Every person, registration reference and source in the exercise is invented training material. The country examples are simplified authored specimens, not official registry forms.
The aim is to put these fragments into a Network while keeping the differences between the sources visible. A shared director will become a connection to inspect. An ownership claim will need a separate justification.
Start with the question, then choose the input
Begin with a narrow question: which of the three companies share a named director in the supplied material? That asks for a particular relationship. It does not ask the application to discover companies, retrieve national records or decide who ultimately controls them.
Choose an input route that fits the fragment in front of you:
| Material | Useful input route | Decision to make |
|---|---|---|
| A short company entry | Manual record | Which values belong in the company's structured fields? |
| Several company rows | CSV entity import | Which column maps to each field, and which rows should enter the Network? |
| A narrative officer notice | AI proposals from pasted text | Which proposed records and relationships does the passage support? |
These are alternative ways to introduce information. Repeating the same company through every route would create a duplicate problem, not additional evidence. In a rehearsal, start each route from its designated checkpoint. In your own work, inspect possible duplicates before adding another record.
1. Enter the short record by hand
For the German specimen, create a Company record and enter its name. Add the legal form, jurisdiction and registration reference in their corresponding fields. Keep the reference as written rather than replacing it with a similar identifier from another country's system.
The useful distinction is between a structured field and the source that supports it. A registration number in an inspector is a value. Its accompanying source tells another reader where that value came from. Add the source details and a citation to the relevant record, with an excerpt or locator when it helps someone check the entry.
Manual entry is particularly useful when a fragment is short or needs careful interpretation. There is no benefit in turning a three-line specimen into a complicated import just to avoid typing it.
2. Map the table before importing it
The French specimen is a semicolon-delimited company list. In the CSV importer, inspect the detected columns and map the company-name column to Name. Select Company as the default kind when the file contains companies. Map legal form, jurisdiction and registration reference to the appropriate fields.
A column that does not fit a registered field can become a visible custom attribute, or you can deliberately ignore it. Check the row preview before applying the import. Empty values, unexpected types and possible duplicates deserve attention while the source table is still in view.
CSV import creates entities. It does not turn an officer's name in an arbitrary column into a director relationship, nor does it automatically create Source or Citation records. Add those deliberately. The selected rows enter together as one undoable action, which makes it possible to reverse an incorrect import without undoing unrelated work.
3. Review the structure suggested from prose
The English training notice names a director:
LB Training — Mara Voss (training person LB-P-001) is a director of LB Training — Merebridge Ltd.
The notice explicitly matches LB-P-001 to the person in the Lindenwerk notice. This authored identifier supports the match; a matching name alone would not. The French record names a different director, Émile Lenoir, for Atelier Rive.
Use the AI checkpoint and paste the full English notice into the integrated AI assistance workflow. Select Merebridge and Mara as the existing context. The workflow uses your own OpenRouter API key. Inspect the result rather than treating generation as a completed research step.
For each proposed director relationship, check both endpoints and the supporting excerpt. Does the officer refer to the intended person? Does the company name match the existing record? Does the passage actually support the relationship type? Reject a proposed ownership relationship if the supplied words do not establish it.
The actual model response can vary. A clear source paragraph does not guarantee a particular proposal, and a proposal that passes validation still needs your judgment. Apply only the items you choose after review.
The same person connects two companies—what follows?
Mara's director relationships to Lindenwerk and Merebridge answer the opening question. They let a reader inspect the person and the supplied source behind each role. Atelier Rive's director is Émile; Merebridge also names Owen Reed.
Those roles do not establish that Lindenwerk owns Merebridge. The fixture retains that tentative ownership assertion as disputed, with the analyst note and correction attached. Ownership remains undetermined; rejecting the inference does not prove that ownership is absent.
The result supports a precise explanation: the supplied material names Mara as a director of Lindenwerk and Merebridge; those roles are represented in the Network; the notices leave ownership undetermined. The analyst writes that explanation. Lage Bureau does not generate a finished due-diligence report in this workflow.
Change the interface, keep the information
Switch the app between English and German, or between dark and light. The interface changes; the company names, entered values and source text remain as supplied. This matters when your records span languages: display preferences should not silently rewrite the material you are inspecting.
Editable Networks stay as unencrypted data in the current browser profile. Use synthetic or non-sensitive material in this prototype, and use validated Network JSON export/import when you need a manual portable copy.
Explore the cross-border company case, or read how the input and review workflow works. The useful test is whether you can explain each connection and recover the source behind it.
About the sources
The company names and quoted notice above were authored for this fictional exercise. No live registry was queried. The related case supplies the training material; it is not evidence about a real company, national filing system or person.
Different countries. Different formats.
Two separate fictional training examples: enter German company details, then map, review and import a French CSV row.
28 seconds · silent · fictional practice data. Includes captions and a descriptive transcript.
Read the descriptive transcript
Different countries. Different formats. — descriptive transcript
This finished 28-second teaser is intentionally silent. There is no speech, music or other audio. The current Lage Bureau logo and the labels “Prototype / Synthetic data” remain visible. All application pixels come from genuine recorded prototype workflows using fictional companies and source text. The German manual-entry example and French CSV example use separate labelled training checkpoints.
00:00–00:02.40 — Opening. The NORTH-EAST SIGNAL logo appears on black with white Archivo text and exact editorial Signal #FF4A17. Text: “Different countries. Different formats.” Supporting text: “Work with the information you have.” The input formats are listed as “Manual entry / Source text / CSV”.
00:02.40–00:05.60 — German company, manual entry. Label: “01 / Germany / Manual entry”. The actual Create entity form is set to Company. The fictional name “LB Training — Lindenwerk GmbH” is entered. Text: “Start with what you know.” Caption: “Enter the company yourself.”
00:05.60–00:08.60 — Local details. The actual Edit entity form contains the manually entered jurisdiction Germany, registration number LB-DE-0001 and legal form GmbH. Text: “Keep the local details.” The caption repeats the exact example values: “Germany / GmbH / LB-DE-0001”. These are fictional training identifiers, not verified registry results.
00:08.60–00:11.60 — Original-language source text. Label: “01 / Germany / Source text”. A close crop of the real Add citation form shows the original German source text pasted into Submitted source text. The visible words include “Status im Übungsdatensatz: aktiv” and a fictional managing-director entry. Text: “Keep the original wording.” Caption: “German source text stays in German.” The source itself identifies the content as a fictional training case. No automatic translation is shown.
00:11.60–00:15.80 — French CSV mapping. The film explicitly switches to “02 / France / CSV import”, with the subline “LB Training — Atelier Rive SAS / separate training example”. The actual mapping interface shows French headers including Dénomination, Juridiction and Immatriculation being assigned to the application's name, jurisdiction and registration-number fields. Text: “Map the French headers.” Caption: “Choose how each column is imported.” The footage uses the supplied French CSV file; it does not query a live registry.
00:15.80–00:19.80 — Review before applying. The real CSV preview shows one source row, one valid company, and no warnings or blocked rows. The record is LB Training — Atelier Rive SAS, with jurisdiction France, legal form SAS, fictional registration LB-FR-0002 and French notes. The Import valid rows control is visible before the actual import. Text: “Review before you apply.” Caption: “One company. Check it before import.”
00:19.80–00:22.80 — Apply the row. The actual recording advances through applying the valid row and selecting the newly imported French company in the graph. Text: “Apply the reviewed row.” Caption: “The company appears in the Network.” This is the recorded CSV training checkpoint, not an implied continuation of the separately filmed German manual-entry checkpoint.
00:22.80–00:25.20 — Import origin and analyst status. The selected French company's actual inspector reads Analyst confidence: Unconfirmed; Created by: CSV import; Sources and citations: 0 citations. Text: “Keep the import origin.” Caption: “CSV origin. Unconfirmed. No citation.” The statement applies to this newly imported record; another record already present in the training checkpoint has its own source and citation.
00:25.20–00:28.00 — End card. Text: “Bring your research together.” “Explore Lage Bureau.” “lagebureau.de”. “A brand of Wehron GmbH.” The current logo and synthetic-prototype labels stay visible.
This film demonstrates manual company entry, original-language source text and explicit CSV mapping, review and import. It makes no claim about automatic translation, document upload/OCR, live country-registry integrations, automatic verification, AI generation, cloud persistence, collaboration or published sharing.
Download transcript (Markdown)