# Principles We Follow

> The practical principles HomeGenomics follows while helping developers and founders learn genomics through real cases and AI.

HomeGenomics needs practical standards we can return to when creating and revising content, not a grand declaration. These are the promises we intend to keep even as the project grows or its contributors change.

## We focus on learning for developers and founders

The primary audience of HomeGenomics is IT developers and founders. We explain genomic and biological data through structures familiar to developers, including code, data models, and reproducible analysis.

Our public contribution does not depend on teaching everyone directly. We help developers and founders avoid overtrusting genomic data or AI analysis, check evidence and limitations, and share sound judgment with their peers and wider society.

## We learn from real cases while sharing methods, not identities

Real cases are valuable because they show where an analysis question came from and how a decision developed. Learning from a case does not require publishing the person's identity or raw data.

- We prefer public or synthetic data.
- When a member contributes a case, we treat internal study use, website publication, source-repository publication, and external presentation as separate choices.
- Each person decides whether their participation, real name, affiliation, account, or contribution history is public.
- No one has to explain a decision to remain private, and anonymous or pseudonymous participation is treated equally.
- Before publication, we consider that information may be impossible to retrieve completely once it has spread.

These principles do not replace consent for a specific case. We will define the concrete consent and data-handling process before accepting a member's case.

## AI output is something to review, not an answer to accept

AI agents can prepare environments, run code, and find material. Their output is trustworthy only when the input data, calculation, actual artifacts, and sources agree, not merely because an AI produced it.

Where possible, an analysis records the input source, tool and database versions, execution conditions, exclusion criteria, and remaining uncertainty. When an agent should not make a choice on its own, it must present the available options and stop.

## We write only what the evidence supports

We do not treat an observed result, a participant's account, a company claim, a media report, and HomeGenomics's interpretation as the same kind of fact. We do not generalize a result from one person or one experiment to others, and we distinguish finding a candidate from demonstrating an effect.

When the evidence is incomplete, we leave the question open instead of filling the gap. We revisit earlier explanations when new evidence becomes available.

## We handle personal and family data only when needed

If public or synthetic data can meet the same learning goal, we do not use personal data. If personal or family data is necessary, we first define why it is needed, where it will be handled, who may access it, and how long it will be retained.

Names, contact details, and family relationships that are not required for the analysis are not stored or shared with it. We also check what leaves the local environment through prompts, logs, and API requests, not only through explicit file uploads.

## We correct errors openly

We record sources and review dates where possible, and reproducible exercises include their inputs and expected results. When we find an error, we aim to record what changed and why instead of silently replacing the conclusion.

We disclose sponsorship or business interests when they relate to the content. Our priority is not promoting a particular product or service, but helping developers judge genomic data and AI results more accurately.

These principles are not finished forever. When practice reveals a gap, we will revise them and record why they changed.

Next: [The case of Sid Sijbrandij](/en/start/case-sid/)