Public Methods

Edition label: 2026-10

This site and its public source repository, github.com/DisclosureWorks/parallax-publications, publish only reviewed, allowlisted, public-facing outputs.

Principles:

• Keep the private application isolated from the public delivery plane.
• Publish from a fixed approved release-output directory only.
• Preserve provenance, hashes, and review status in the public artifact set.

Non-goals:

• No private corpus mounting.
• No wiki indexing.
• No hidden editorial state.

Legal and privacy review

Applies to Parallax Publications and the Parallax Research Atlas.

1. Why this exists rather than relying on removal

The removals policy § 6 states the limit honestly: under CC BY 4.0, removal does not retract copies. That makes pre-publication review the real protection, and the removal route a backstop for what review misses. This policy is where the protection actually lives.

No review, rights, privacy, provenance, or security gate is waived to meet a publication date. Where a date and this policy conflict, the component is descoped.

2. Mandatory review triggers

A candidate item may not enter a public edition until it has been reviewed against this policy when any of the following applies:

1. It makes a factual claim about a named living private individual.
2. It makes an adverse factual claim about a named living public figure — meaning any claim of wrongdoing, misconduct, or criminal conduct.
3. It touches sealed, grand-jury, or Rule 6(e)-adjacent material, or analyses it.
4. It is victim- or minor-adjacent — drawn from a document, case file, or production known to contain such material.
5. It contains or derives from recovered-redaction content (the redaction policy § 4).
6. The publication gate flagged it — restricted path, missing sensitivity metadata, PII pattern, wrong destination, or non-allowlisted item.
7. It is court-record-derived and has not previously appeared in a published edition.

Where the review load cannot be carried before a release, the affected findings are descoped rather than published unreviewed.

3. What review checks

• Accuracy — the claim is supported by the cited primary source, read directly rather than through a secondary characterisation.
• Provenance — the cited document resolves, its hash matches, and the citation points to the primary rather than to an aggregator.
• Privacy — the standing redactions in the redaction policy § 2 have been applied.
• Characterisation — allegations are described as allegations; charges as charges; findings as findings. A conflation here is the most likely source of a defensible-claim becoming an indefensible one.
• Necessity — where sensitive material is included, that its inclusion is necessary to the finding rather than incidental colour.

4. Recording review

Review state is not a memory or a conversation. Every registry item carries review state, accountable owner, provenance, and exact content hash. The gate is fail-closed: an item with no sensitivity or approval metadata is refused, not admitted.

Source material does not always carry sensitivity metadata. Fail-closed is what makes that safe: absence of metadata is treated as “not cleared”, never as “nothing to clear”.

Because the hash is recorded, review attaches to an exact version. A reviewed item that is subsequently edited is unreviewed again. There is no such thing as inheriting an approval across a content change.

5. Who reviews

The publication owner. The single-approver posture is explicit: the person who decides an item has cleared review is the same person who signs that the edition may publish. There is no independent second reader.

This is the weakest point in the first edition’s controls and should be read as such. It is acceptable for a first edition only because the technical gates are fail-closed and independently tested, so the human is not the only thing standing between restricted material and a public URL. It should be revisited after the first edition.

6. Outside counsel

This policy is not a substitute for legal advice, and its author is not a lawyer. Four matters warrant legal counsel: defamation exposure on adverse claims about named living people; the Rule 6(e) line; whether EU or UK data protection obligations attach to publishing personal data from US court records on a public domain; and whether a DMCA designated agent is required.