# PDR 002: Reimplement from documented public sources

- HTML version: https://robbiepalmer.me/projects/pathologize/pdrs/002-public-source-provenance-boundary
- Project: Pathologize
- Status: Proposed
- Opened: 2026-11-06

## Context

The project owner previously worked on commercial computational-pathology systems. General experience can help
identify useful research questions, but the earlier employer may own code, data, models, annotations, inventions
and confidential combinations of methods created during that employment.

Public papers and datasets make independent research possible. They do not publish every internal implementation
detail, grant rights to old work, or make a public dataset unrestricted for redistribution.

## Proposed decision

Implement Pathologize only from cited public papers, appropriately licensed public software, public datasets and
new project-authored work. Record the source and licence for every external dataset, model and substantial method.

Do not access, copy, infer or reconstruct former-employer code, data, weights, annotations, configurations,
thresholds, experiment histories, internal diagrams, unpublished failures or private product knowledge. Do not
publish claims about a former product unless an authorised public source supports the exact claim.

When a remembered technique has no public source, exclude it until independent public evidence supports the
method. Review former employment and confidentiality agreements before commercialising the result. Treat patent
clearance as a separate question from source-code independence.

## Consequences

The repository gains an auditable explanation for where its work came from. Public citations and new code reduce
disclosure and ownership risk while making the research easier to reproduce.

The owner cannot make this a perfect legal clean room because prior knowledge cannot be removed. Some apparently
obvious improvements will be omitted or require legal advice. Public provenance also adds documentation and
licence checks before experimentation begins.

## Evaluation

Accept the proposal only if a review can account for every dataset, model and nontrivial implementation source
without relying on private employment material. Pause publication or commercialisation when that account is
incomplete or an old agreement creates unresolved ownership or confidentiality risk.

## Related records

- Project: [Pathologize](https://robbiepalmer.me/projects/pathologize.md)
- Project: [Genomic Prediction from Histology](https://robbiepalmer.me/projects/genomic-prediction.md)

---

Markdown index of this site: https://robbiepalmer.me/llms.txt