Licensing Laboratory Data for AI: What to Inventory Beyond the Final Test Result

A laboratory data inventory should explain how a sample became a released result. Start with the links between samples, methods, instrument runs, quality controls, reviews and reports, not a headline file count. This guide provides a worksheet for describing that archive before a licensing discussion.

It is written for commercial testing laboratories, LIMS owners and technical managers. Clinical laboratory records can require additional patient-data review; the EHR inventory guide addresses that different setting. An inventory is preparation, not permission to disclose records or a promise that someone will buy them.

What counts as laboratory data?

Laboratory data can include raw instrument outputs, processed measurements, sample descriptions, method versions, quality-control results and the final report. These are different layers. A PDF certificate may summarize a result without preserving the signal, settings or review history that produced it.

Begin with one bounded archive: a test family, a source system and a period your team understands. Name the sample matrices, such as water, soil or manufactured materials, without identifying clients or confidential products. Our laboratory datasets overview describes the broader record families.

Do not assume that a LIMS export contains the whole history. Instrument software, shared folders and quality-management systems may hold essential context. Record where each part lives and which connections have actually been checked.

Use a laboratory archive manifest

Complete this worksheet internally. A version shared for an initial enquiry should describe fields and coverage, not contain real sample identifiers, client information or instrument files.

Inventory item	What to establish internally	What to describe initially
Archive boundary	Lab sites, systems, test families and dates included	Broad scope and known exclusions
Counting units	Samples, tests, runs, results, reports and files separately	Approximate scale with the counting rule
Sample context	Matrix, preparation, sampling conditions and chain of custody, where retained	Available context and gaps
Method	Method identifier, SOP version and changes over time	Whether version history is traceable
Instrument output	Native files, exported signals and processing settings	Formats, software dependencies and availability
Quality controls	Blanks, standards, controls, calibration and batch relationships	Which control records link to each run
Result meaning	Value, unit, qualifiers, limits and status	Definitions that accompany measurements
Review history	Analyst review, amendments, retests and release events	Whether decisions and versions are retained
Linkage	Stable internal references across systems	Verified links versus assumptions
Coverage	Missing periods, failed exports, retired instruments and migrations	Limitations and how they were identified
Authority	Client terms, third-party rights and institutional approval route	What review is complete and what remains open

This is an original preparation worksheet, not an accreditation checklist. ISO’s overview of ISO/IEC 17025 describes a standard concerning laboratory competence, impartiality and consistent operation. Accreditation does not establish that a particular archive may be licensed for a new purpose.

Count samples, tests and files separately

A sample may undergo several tests. A test may involve repeated runs. A run may produce several files, and a report may include multiple results. Reporting these as interchangeable counts makes an archive look larger or more complete than it is.

State the unit beside every estimate and retain the query or counting method internally. Distinguish unique records from repeated exports and revised reports. If part of the archive has not been examined, label the estimate accordingly rather than projecting a convenient folder count across every system.

For a proposed AI task, the useful example could be a spectrum linked to its interpretation, a complete analytical batch or a result-review sequence. That unit should be agreed later. It should not silently replace the archive’s original counting units.

Preserve the route from raw signal to result

A raw signal and a released measurement answer different questions. A chromatogram can preserve instrument response over time; a final report states the result the laboratory released. Neither should be described as a substitute for all the other records.

Ask whether your systems retain processing methods, integration changes, dilution factors and the relationship between a native file and its export. Document software and format versions where known. Do not describe a proprietary binary file as immediately usable simply because it can be copied.

Keep the original representation distinct from any later conversion. Record the transformation, its version and any fields it drops. Do not overwrite original units or qualifiers just to make a spreadsheet look consistent. The FAIR principles emphasize rich metadata, provenance and usage conditions for reuse. They are a useful reference, not a requirement to make confidential data public.

Make units, limits and method changes explicit

A number without a unit, method and qualifier can be misleading. Document the unit as recorded, whether a value is calculated or directly reported, and the meaning of flags such as not detected or below a reporting limit. A missing value, a zero and a qualified result are not interchangeable.

Describe any changes in method, instrument, preparation or reporting convention across the archive. A consistent column name does not prove that every value was produced in the same way. If comparability has not been assessed, say so.

Hypothetical example: a testing laboratory retains reports across two method versions. The later version records an additional qualifier, while older exports omit it. The inventory should identify the change and the affected coverage. It should not fill older blanks with an assumed result. This example is illustrative and does not describe a DataNexx supplier or available dataset.

Keep failed runs, retests and amendments distinguishable

A repeated measurement may be a planned replicate, a rerun after a control failure, a confirmation test or a correction. Those explanations have different meanings. Preserve the reason where it was recorded and mark it unknown where it was not.

Retain the relationship between the original and later records. A final accepted result should not erase the earlier run from an internal inventory, but that does not mean every record belongs in a licensed extract. Selection depends on the agreed task, technical suitability and permissions.

Keep quality-control status separate from product compliance or safety conclusions. A control passing within a laboratory process is not a universal guarantee about the sample, a product or an AI system trained on related records.

Check the links, not just the presence of fields

Ask a technical colleague to trace a small internal selection from report to sample, method version, run and review. Note links that are deterministic, inferred or unavailable. Do this inside approved systems rather than emailing records around for convenience.

Investigate broken references, duplicate filenames, overwritten exports and references that changed during migration. Count examples with the required linked components separately from examples where only the final report survives. Your scientific AI data discussion can then focus on the actual evidence an archive preserves.

Separate a useful description from a permitted transfer

Possessing files or having administrator access does not establish rights to license them. Customer testing agreements, confidential formulations, staff information and third-party software or reference materials may restrict a proposed use. Removing customer names alone does not resolve all these issues.

Before any sample transfer, identify the proposed purpose, recipients, countries, allowed uses and relevant approval route. Use a suitable secure transfer arrangement and an agreed scope. This is general information, not legal advice; institutional and legal review should address the actual records and destinations. See our data licensing principles for the distinction between provenance and authority.

What should a first enquiry contain?

A useful opening description names your laboratory type and country, broad test families, approximate period, counting units, formats and the links your team has confirmed. State whether raw outputs, methods, QC and review history exist. Include limitations rather than claiming that the archive is complete.

You do not need to send raw data to start. Describe your laboratory archive with a high-level summary only. DataNexx can use that description to discuss potential fit as an intermediary; an assessment does not guarantee licensing permission, a buyer or a transaction.