Ethics, Integrity & Data-Privacy Guidelines

Report what you found, including the parts that did not work. Negative and inconclusive results are results, and a programme that only hears about successes learns nothing.

Do not fabricate data, alter a figure to make an effect look cleaner, or omit results that contradict the conclusion. Do not present someone else's text, code, figures or ideas as your own - cite them.

Keep records as you go: the version of the data, the parameters, the code and the date. If a result cannot be reproduced from your own records, it is not yet a result.

Authorship follows contribution. Agree the author list and the order with your mentor before a manuscript is drafted, not after.

Everything the platform returns is a prediction with a confidence attached, not a measurement. Treat it as a hypothesis to test, and say so in anything you write.

Check a claim against the underlying record before you rely on it. Every score in the product opens into the evidence that produced it - that is what it is for.

Declare where model output, generated text or AI assistance was used in your work. Never present generated content as an experimental observation, a citation you have not read, or a result you did not obtain.

Do not attempt to extract, copy or reconstruct the platform's models, training data or underlying datasets, and do not use programme access for work outside your approved project.

Work involving human participants, human samples or identifiable human data needs approval from the appropriate ethics committee or institutional review board before it starts. Hold on to the approval reference - you will be asked for it.

Data must reach us de-identified. Strip direct identifiers at source, and do not upload names, contact details, identification numbers, full dates of birth, facial images or free-text notes that identify someone.

Never attempt to re-identify an individual in any dataset, and never link datasets in a way that would make re-identification possible.

Special-category data - health records, genetic and biometric data - carries the highest bar. If you are unsure whether a dataset qualifies, ask before you upload it, not after.

Collect the minimum you need, keep it only while you need it, and use it only for the purpose it was collected and consented for.

Keep programme data inside the systems provided for it. Do not copy it to personal devices, personal accounts or consumer file-sharing services, and do not paste it into third-party tools - including public AI assistants.

Your credentials are yours alone. Do not share them, and do not give anyone access on your behalf; access is granted per person and per project.

Encrypt anything you must hold locally, delete it when the project ends, and confirm the deletion if you are asked to.

If data is lost, exposed or accessed by someone who should not have it, report it to info@prognica.com as soon as you become aware - within twenty-four hours, and before you attempt to fix it yourself.

Treat unpublished methods, datasets, model behaviour, roadmap details and anything shared by another fellow as confidential unless you are told otherwise in writing.

Share your intended publication, preprint, poster or talk with the programme before it goes out, so confidential material and unfiled intellectual property can be checked. The review is for that purpose only - it is not a veto over your findings or your conclusions.

Ownership of what you create during the programme is set out in the fellowship agreement. Nothing on this page changes it.

Laboratory work follows the biosafety and biosecurity rules of the institution where it is performed, and the fellow performing it is responsible for holding the required training.

Some findings can be misused. If your work could plausibly assist in causing harm - to people, to a population, or to a system others depend on - raise it with your mentor and the programme before you publish or share it, and expect that conversation to be taken seriously rather than treated as an obstacle.

Do not pursue work whose predominant purpose is to cause harm, and do not use programme resources for it.

Disclose anything that could reasonably look like it might influence your work: funding, employment, consulting, equity, advisory roles, patents, and close personal relationships with people involved in reviewing or supervising you.

A disclosed conflict is normally managed and the work continues. An undisclosed one is the problem.

The recognition and case-study consent on the application form is optional, and the application is assessed the same either way.

If you give it, we may name you and describe your project in programme materials. You can withdraw that consent at any time by writing to info@prognica.com; material already printed or published cannot be recalled, but nothing further will be produced.

If you see something that looks like a breach of these guidelines - yours or anyone else's - report it to info@prognica.com. Concerns can be raised in confidence.

Nobody who raises a concern in good faith will be penalised for it, whatever the outcome of the review. Retaliation against someone who reports a concern is itself a breach.

Reports are reviewed by the programme. Where a concern is substantiated, the response is proportionate: it may be guidance, a correction to the published record, suspension of platform access, or withdrawal from the programme. Serious matters may be referred to your institution or to a regulator where we are required to do so.

These guidelines are reviewed at the start of each cohort and whenever the programme changes materially. The date at the top of this page is the version in force.

Substantive changes are notified to current fellows by email. Questions about anything on this page go to info@prognica.com.