Skip to content

“Private” is a claim you support with an evidence chain

Private data is not simply data kept behind a login today. A licensing review needs evidence about how the collection was created, who could access it, whether any part was public, what personal or confidential material it contains, and whether earlier commitments limit its use.

This checklist is for preparation only. It cannot prove that a repository was never exposed, that every contributor granted rights, or that a dataset was not used in model training. General information, not legal advice.

Updated

On this page

Review five surfaces

General information, not legal advice.

Public exposure. Search the collection’s known names, package markers, domains, repository references and published releases. Check public mirrors and forks where relevant. Record the date, query or method, and the result. A clean search result means “not found by this method,” not “never public.”

Full history. Review the complete history that will be offered, not only the current files. The official secret-scanning guidance says its scan covers the entire Git history on all branches for supported hardcoded credentials GitHub Docs, (opens in a new tab). That supports a useful technical principle: current-file review alone is not enough. It does not prove that every secret or every sensitive field was detected.

Contributors and third parties. List employees, contractors, clients, open-source components, data suppliers and customer material. A private repository can still contain code written under a client agreement or a component with obligations that do not match the proposed license.

Personal and confidential fields. Look for names, email addresses, phone numbers, tokens, credentials, customer records, support messages, HR material, legal privilege and internal identifiers. Mark what is excluded, masked or escalated. Do not publish a sample that includes a real value while demonstrating a scan.

Model-generation signals. Record known generated files, assistant metadata and the limits of the check. Do not claim that a scan can identify every model-written line or guarantee that training data has never appeared in an evaluation set.

The evidence worksheet

General information, not legal advice.

The evidence worksheet (table)
SurfaceEvidence referenceResultAction
Public repositories and mirrorsDated search logFound, not found, or incompleteExclude, investigate, or document the limit
Full history and branchesScan report and clone stateComplete, shallow, or unknownObtain full history or narrow scope
Contributors and clientsAssignments and contractsOwned, permission needed, or unresolvedRequest consent or exclude
Dependencies and componentsLockfile or SBOM reviewLicense identified or unknownDisclose, remove, or ask your own lawyer
Secrets and personal dataScan report plus manual sample reviewFinding count and scopeRotate, redact, exclude, or re-scan
Prior licenses and termsContract registerNo conflict, conflict, or unknownCarve out or get legal advice
Model-generation signalsTool output and owner statementSignal found or check unavailableDescribe limitation, never overclaim

Worked fictional example

General information, not legal advice.

Suppose a fictional team has a private repository with complete local history and a public package published years ago. The current files contain no obvious credentials, but the history has an old token and one folder includes client integrations.

Worked fictional example (table)
FindingSafe conclusionNext step
Public package markerSome component was publicExclude the package and inspect forks and releases
Old token in historyA credential was present in the reviewed historyRotate it, remove the affected history where appropriate, and record the incident review
Client integration folderTitle is not established by repository accessFind the client permission or leave it out
Full-history cloneThe scan covered the intended historyKeep the clone-state evidence with the report

The safe output is a narrower candidate scope, not a “private” badge for the entire repository. A review can proceed on the internal service after the client folder and public package are excluded, subject to title review and your own lawyer’s advice.

Rules for a safe sample

  • Run checks on the owner’s machine or the approved isolated process. Do not upload a private archive to complete this worksheet.
  • Keep scan output separate from code, paths, emails and credentials. Share only the bounded report the route requires.
  • Record the source revision, scan date, clone state and exclusions.
  • Treat a tool result as evidence of what it checked. Keep unresolved areas visible.
  • Re-run after remediation. A redaction or history rewrite changes the evidence set.

Rightmark Repo Check is the route-specific technical starting point for private code. The Trust page explains the ownership pack and approval gate. For workplace archives, use the operational data page to define systems and exclusions.

This checklist does not establish that data is private, non-infringing, uncontaminated, accepted by Rightmark or worth a particular amount. It makes the next review question visible.

Sources

  1. Information Commissioner’s Office: Data protection by design and by default (opens in a new tab)
  2. GitHub Docs: Secret scanning (opens in a new tab)

Repo Check: value your repository without sharing your code

One command on your own machine reads a repository and its git history. You share a file of counts and fixed labels, never code, and see an indicative range, which route fits and what to fix first.