Source documents
Archives
Coherent document archives with documented chain of custody.
PDF · EML · XLSX · DOCX
The General Data Company
Real operational data. From real European companies.
Operational data from European companies, prepared for AI training and evaluation. Anonymized, rights cleared, and unavailable on the open web.
Reality is the dataset.
0%
synthetic. Every case happened
100%
anonymized before it leaves the company
1JSON
per case: source, process, verified outcome
Source material
Records inside real companies
Company archiveCorrespondence and files
Plant recordsProcess and inspection logs
Office recordsOrders and invoices
Each source stays linked to its records.
Illustrative workflow
From exception to resolution
Routine sequenceThe expected steps
Late deliveryAn exception in the process
Resolution recordedWhat actually worked
The case connects the problem, the decision and the result.
Connected records
From source files to case.json
Source filesWorking documents
DecisionsSteps taken
Verified resultRecorded outcome
case.jsonOne linked trajectory
Source documents, actions and outcome stay together.
Privacy transformation
Same case. Protected identities.
Consignment recordCMR · 04173
Kühlfracht Hessen GmbH
Metro C&C, Filiale Kassel
TK Gemüse, 12 Pal.
−18 °C · ok
M. Yilmaz
KS KF 4471
06:40
14:15 (+3h, Stau A7)
Illustrative recordOperational fields retained
- Detect
- Replace
- Verify
Original values
Rights and delivery
From mandate to usage license
Usage licenseTraining or evaluation
Cleared corpusTransformed records
Written mandateDefined source scope
The same corpus is licensed for defined uses.
Agent evaluation
A task with a verifiable result
Task contextStarting records
Agent actionChoose the next step
Task scoreAgainst a verified outcome
The recorded outcome provides a basis for scoring.
01 · Hidden work
Real operations.
Recorded every day.
Plants, warehouses and offices leave records of decisions and exceptions. We structure that work from ERPs, shared drives and paper.
- SourceERP, shared drives, paper
- AccessNever without written authorization
02 · Reality
The exception
is the data.
A late delivery. A rejected batch. A workaround that succeeded. Real cases connect each decision to its verified outcome.
- CaseException and resolution
- ResultVerified outcome
03 · What we take
Documents.
Decisions. Outcomes.
We connect the working documents, decisions and result into one trajectory per case, with the source archive underneath.
- TrajectoryOne JSON per case
- ArchiveDocuments labeled per field
04 · Method
Replaced,
not redacted.
Personal values are replaced consistently across each case. The mapping is deleted, and the transformed documents are checked and labeled.
- ContextCross document links retained
- ReviewTwo annotators, checked
05 · Rights
Rights at
every stage.
Every corpus carries its own chain of title, from the written mandate at the source to the license that leaves the building.
License
Licensed, never sold.
Defined usage. No resale.
Transformation
The transformed output.
Structured, anonymized and labeled.
Supply
Authorization at source.
Written scope before a file moves.
06 · Environment
From trace
to environment.
Real trajectories become tasks with a known outcome. Evaluation sets come first, followed by environments for training and scoring agents.
- TasksNext step, exceptions, grounded QA
- DeliveryManifest, privacy record, license
That is the asset.
Here is how you buy it.
Data for AI teams
Three ways to work with real data.
Historical archives and active workflows, delivered as source material, structured data or evaluation tasks.
Structured data
Structured extracts
Anonymized, entity resolved extracts, validated against your schema.
JSONL · Parquet
Model evaluation
Evaluation sets
Unpublished tasks for extraction, reasoning and agent workflows.
Task + rubric pairs
Evaluation and clearance preview
Illustrative interface and scores. Not measured benchmark results.
Held out document tasks · Score
Claude Opus 50.742
GPT 5.6 Sol0.718
Gemini 3.1 Pro0.655
DeepSeek V4 Pro0.588
Console · Clearance pipelineCorpus LCK · Rev 2.1
DocTypeEntitiesStatus
D-88121Email · thread64Cleared
D-88123Contract · annex203Redacting
D-88124Ledger · export41Redacting
Throughput · docs / day
Entities redacted1,284,507
Sources
Where it comes from.
Authorized at source. Reviewed per corpus. Nothing is scraped.
Insolvency estates
Document estates from wound-down European companies, which administrators release only under a written mandate.
Workflow capture
Expert work recorded with consent as the work happens.
Personal documents
A contributor app for on device anonymization. Not yet released.
Bespoke corpora
Professional datasets assembled against a defined brief.
Buyers
For AI teams.
Licensed for training and evaluation, with a manifest, privacy report and defined terms.
Rights
Written chain of title.
Privacy
Batch verification and reporting.
Exclusivity
Exclusivity windows by agreement.
Delivery
Manifest with every file.
Position
Data the internet has never seen.
Method
From estate to dataset.
1.0
Acquire
A written mandate before a single file moves.
2.0
Clear
Ownership and third party rights reviewed.
3.0
Anonymize
Identifiers transformed and checked per batch.
4.0
Deliver
Corpus, manifest and license.
A few practical questions.
What is The General Data Company?
We source operational data from European companies and license it to AI teams for training and evaluation.
What does The General Data Company sell?
Archives, structured extracts, evaluation sets and environments built from real workflows.
Is the data GDPR compliant?
Each corpus undergoes rights and privacy review, with verification before delivery.
Can we license data or environments exclusively?
Yes. Corpus and task collection exclusivity is agreed per license.
Contact
Get in touch.
Tell us your target domain, data format and evaluation needs. We will discuss the relevant archives, extracts or task collections.
hello@thegeneraldata.com