Unpublished work is competitively sensitive. The product is built as if it were ours to lose.
Ownership follows your institution's policy
US universities own research data produced with their resources and the principal investigator is its steward. Methods contracts on exactly those terms: the researcher is the customer, the institution's data stays the institution's, and nothing leaves your workspace without an explicit share action.
No training on your protocols, no retention by default
Model calls run under commercial terms with zero data retention where available. Your protocols are never used to train models. The model families that require retention are not used.
Confidential by construction
Your agent is instructed not to ask about results, hypotheses or unpublished findings beyond what the procedure needs, and not to repeat a confidential mention back.
Wrong is visible, not hidden
A missing value is written as Not stated. Uncertainty is written in your words. A correction you make at sign-off is versioned and, if a protocol was already shared, every recipient is notified of the change.
You are the author
Nature, Science, Cell Press and PLOS all hold human authors fully responsible for AI-assisted text and none allow AI authorship. Methods is a drafting and elicitation tool. Your name goes on the protocol because you approved every line.
Export is trivial on purpose
A protocol should never live only here. One click sends it to your notebook, your repository, or your funder's deposition service.
on your computer.
At every tier.
Your workspace holds what we create together: the record, the workflow inventory, the templates, the standardization map and the results you choose to save. Versioned, exportable in full, deletable in full.
Scripts are tested on synthetic data and handed to you to run yourself. We test the scripts and give you the output to make sure they work, but we never hold, store, or have any of your own raw data.
Approved workflows run on your institution's cloud account or ours, one approval at a time. The raw files stay in the account you chose.
Straight answers
Who can see my protocol?
Only members of your lab workspace, and only the people you add. Value Stream Science staff do not read protocols except with your written permission for support.
Does my protocol train the model?
No. Model calls run under commercial terms that exclude training on customer content, with zero data retention where the provider offers it.
What if the agent gets something wrong?
It cannot invent a value: anything you did not say is marked not stated. If a wrong claim still reaches an export, the correction is versioned and every recipient of that protocol is notified.
Does this create an invention disclosure?
A captured protocol is not normally patentable, but if it reveals a novel method developed under federal funding it may evidence a disclosable invention under your institution's Bayh-Dole obligations. Export is one click for exactly that reason: the protocol should also live in your institution's own systems.
Does the agent keep a copy of my documents?
The index of your lab's folders is built on your own computer. For each answer, your agent reads only the passages it needs and cites them. Nothing is copied to your lab's workspace unless your PI chooses to share the index with the lab, and a shared index is visible only to your lab and deleted with it. Your documents are never used to train a model, and raw instrument data is never indexed, at any tier.
Do you ever store my raw data?
No, at any tier. Tier 2 scripts run on your own computer against your own files; what comes back to the workspace is the result you chose to save, such as a table, a figure or a metric. Tier 3 runs approved workflows on your institution's cloud account or ours, and the raw files stay in the account you chose.
Can I delete everything?
Yes. Deleting a lab workspace removes its records, transcripts, templates, saved results and history from our systems.