Yes, including the work that failed. AI buyers ask chemistry and materials companies for dated research records that show how the work was done: the question, the evidence, the experiment chosen, the result and the decision that followed. Failed, null and inconclusive runs are wanted with their outcome labels. What decides whether a set can be licensed is ownership: sponsor and customer contracts, grant terms, trade secrets and export controls.
- Buyers want the reasoning behind the work, not the polished paper.
- Failed, null and inconclusive runs are wanted, with outcome labels kept.
- Records made at the time beat retrospectives written later.
- Customer-owned and partner-owned data is not yours to license.
What R&D records do AI buyers ask for?
Records that show how scientific work is done. A public listing for chemistry and materials R&D records (one buyer's public listing, read 15 November 2026) says: "We want the reasoning behind the work, not just the scientific facts or the polished papers."
The unit it describes is the linked research episode: "Each one connects the question, the evidence available at the time, the competing explanations, the experiments or analyses chosen, the results and the decisions that followed." It adds that "Records made at the time are preferred over looking back afterwards."
The listing sorts what it collects into seven groups.
| Group | Records the listing names |
|---|---|
| Hypotheses and research decisions | Research proposals, dated lab notebooks, experiment plans, decision logs |
| Experimental design and protocols | Versioned protocols, controls, randomization or blinding, sample-size rationale, equipment settings, deviations |
| Raw observations | Instrument outputs, images, assay results, sensor or time-series data and simulation outputs, linked to sample IDs, units, calibration and batches |
| Analysis and reproducibility | Notebooks, code, dependencies, preprocessing, statistical models, version history |
| Failures and troubleshooting | Failed or null experiments, anomalous results, investigation logs, abandoned approaches, linked reruns |
| Critique and replication | Internal technical reviews, competing interpretations, replication attempts |
| Research-program history | Project registries, initial forecasts, milestones, resource constraints, stop/go decisions |
Are failed and null experiments worth anything?
Yes. The listing says so directly: "Failed, null, inconclusive, replicated and successful work are all wanted, with outcome labels kept."
The reason is simple. A paper reports what worked. The notebook holds everything that was tried, in order, with the reasons for stopping. A model trained only on published results never sees a wrong turn or how a chemist recognized one. Your records of the formulation that gelled, the coating that delaminated and the cell that faded early are the part no publisher has.
Two conditions make failed work usable. The outcome has to be labelled, so a failed run is not mistaken for a success. And the run has to be linked to what came next: the investigation, the change made and the rerun.
Which labs and R&D companies qualify?
The R&D records listing puts "Materials, chemistry and batteries" first among its priority domains and also names "Industrial engineering and process R&D", with agritech, environmental and climate R&D among the industries it covers.
A second listing from the same source covers semiconductor and materials companies, and names "materials and battery R&D labs" (one buyer's public listing, read 15 November 2026). Its bar is specific: "Run or build records the company owns, with recipe or process settings linked to measured performance", "1,000+ runs, builds, wafers or experiments", "Every build kept, failed ones included, with dates" and "Data that hasn't been published". It gives a size of 10 to 2,000 employees and says "Labs and startups that have wound down qualify."
It also says what does not fit: "Data owned by a customer or foundry partner", "Published papers and datasheets only" and "Licensed third-party IP". Biotech and contract labs are covered in the guide for biotechs, CROs and contract labs, and process data from fabs in the semiconductor process data guide.
What makes a set of R&D records worth more or less
Linkage decides most of it. A result nobody can tie to a recipe, a sample and a date is a number.
| Raises value | Lowers value |
|---|---|
| Every run tied to a sample ID, recipe, date and measured result | Results in spreadsheets with no link to how the sample was made |
| Failed and repeat runs kept and labelled | Only the runs that went into a report |
| Notebook entries written on the day | Summaries written at project close |
| Stop and go decisions recorded with their reasons | Projects that just end in the record |
| Unpublished work | Data already in papers, patents or datasheets |
| Clear ownership across the whole set | Sponsor, customer and partner data mixed in |
How notebook and sample systems differ, and where each record usually sits, is covered in ELN vs LIMS. Records kept under good laboratory practice have their own rules, covered in the 21 CFR part 58 guide.
Who owns R&D records made under a sponsor, grant or customer contract?
The contract decides, and R&D companies usually have several kinds. Work a customer paid for under a development or toll agreement often belongs to the customer, and the listing treats customer-owned data as not a fit. Joint development agreements split ownership by field or by who invented what. Read each one before counting its records as yours.
Federal grants follow a written rule. "Title to intangible property acquired under a Federal award vests upon acquisition in the recipient or subrecipient", but the recipient "must not encumber the property without the approval of the Federal agency or pass-through entity", and the government keeps the right to "Obtain, reproduce, publish, or otherwise use the data produced under a Federal award" (2 CFR 200.315). The award terms can add more, so read the award.
Trade secrets need a deliberate choice. Federal law protects information as a trade secret only if "the owner thereof has taken reasonable measures to keep such information secret" and it "derives independent economic value, actual or potential, from not being generally known" (18 U.S.C. 1839). A licence is a disclosure, so it needs confidentiality, access limits and a bar on publication. Live formulations and current process windows often stay out altogether.
Export controls apply to technical data, not only to goods. Under the Export Administration Regulations an export includes "An actual shipment or transmission out of the United States", and releasing technology "to a foreign person in the United States" is a deemed export (15 CFR 734.13). Published information is outside the rules when it "has been made available to the public without restrictions upon its further dissemination" (15 CFR 734.7). Unpublished process records are the opposite case, so classify them before a sample leaves. This page is general information, not legal advice.
Example: one research episode, assembled
A fictional coatings company pulls together one episode from a cancelled adhesion project. Nothing new is written. The work is finding the records and linking them.
| Step | Record | Where it was |
|---|---|---|
| Question | Project proposal: can primer B replace primer A on aluminium? | Shared drive |
| Evidence at the time | Two supplier datasheets and a 2019 internal test report | Shared drive, notebook reference |
| Competing explanations | Notebook entry: adhesion loss from surface prep or from cure schedule | Electronic lab notebook |
| Experiment chosen | Plan for 24 panels across three cure schedules, with controls | Electronic lab notebook |
| Raw result | Pull-off values per panel with panel IDs and instrument calibration date | Instrument export, sample system |
| Failure | 18 of 24 panels failed after humidity ageing; photos attached | Electronic lab notebook |
| Decision | Review minutes: stop, reason recorded, two reruns listed | Project registry |
Fictional example. Seven linked records from four systems make one episode. A lab with a few hundred of these has what the listing describes.
What a licence usually allows and how long it takes
A licence for R&D records usually lets the buyer use them to train and test its own models under confidentiality. It usually bars publication, resale and any attempt to pass the records to a competitor. Because the records are secret by nature, the confidentiality and security terms carry more weight than in most data licences.
Ask what the buyer may do with outputs, whether named compounds, customers and suppliers are masked, and whether you keep every right to use and patent your own work. There is no standard timetable. The order is usually the same: a written description of the archive, a small sample, a rights review, agreed terms, then delivery. The rights review sets the pace. For R&D sets the slow steps are contract review and export classification.
Common mistakes with R&D records
Most of them remove the very thing buyers ask for.
- Cleaning out failed runs before the count.
- Writing tidy retrospectives and offering those in place of the dated originals.
- Exporting results without the sample IDs that link them to recipes.
- Counting customer-funded and partner-owned projects as your own.
- Skipping export classification because "it is only data".
- Sending a sample that includes a live formulation.
- Letting notebook and instrument software subscriptions lapse when a programme is cancelled, before the records are exported.
R&D records as a data asset
Two published company ranges apply here (one buyer's public listings, read 15 November 2026). The listing for chemistry and materials R&D records runs from $10,000 to $300,000. The listing for semiconductor and materials companies with run records runs from $100,000 to $750,000. Read both as published ceilings, not as typical outcomes: they assume owned, unpublished, well-linked records.
A cancelled programme or a closed lab still has the records, and they can be licensed by whoever holds title to them. The first job is to export them while the systems still run. Broader R&D record types are covered in the R&D experiment records guide and regulated manufacturing records in the GMP batch records guide. Origin Data Partners is an independent data partner that helps companies get their data bought. A buyer may pay Origin when a deal closes.
Does your company hold records like these? See what your data is worth.
Common questions
Are failed experiments worth anything to AI buyers?
Yes. A public listing read on 15 November 2026 says failed, null, inconclusive, replicated and successful work are all wanted, with outcome labels kept. Failed runs show reasoning that published papers leave out.
What chemistry and materials R&D records do AI buyers ask for?
Dated records that link a question to the evidence, the experiment chosen, the raw result and the decision that followed. The listing groups them as hypotheses and decisions, protocols, raw observations, analysis, failures, critique and replication, and programme history.
Do paper lab notebooks count?
The listing asks for dated lab notebooks and does not exclude paper. Paper has to be scanned with page order and dates intact, and the entries need to link to sample IDs and results, so expect a buyer to ask for a scanned sample first.
Who owns R&D data produced under a federal grant?
Under 2 CFR 200.315 title to intangible property acquired under a federal award vests in the recipient, the recipient must not encumber it without approval, and the government keeps the right to use the data. The award terms can add conditions.
Can a lab that has closed still license its records?
Yes, if someone holds title to the records and has authority to sign. One listing says labs and startups that have wound down qualify. The records have to be exported before notebook and instrument systems are switched off.
Does licensing R&D records destroy trade secret protection?
It does not have to. Federal law asks whether the owner took reasonable measures to keep the information secret, so a licence needs confidentiality, access limits and a bar on publication. Many companies also hold back live formulations.
