As of May 2026—the snapshot below describes posture at publication; present-tense “today” in the diagram edge labels that snapshot.
This article is about what must not appear on the Labs site and how reviewers enforce that boundary before anything ships. The site-wide framing—why TB publishes articles instead of open-sourcing Trunk—is on the About this Labs site landing; here the focus is the controls.
What we protect
Material falls into a few broad classes. Confidential or restricted content—portfolio and deal detail, third-party information received in confidence, credentials, and source from private repositories—must never appear in MagnetMagpie git or on the public site. Internal operational detail from Trunk may inform high-level articles once synthesised for this repo, but combinations that leak context still matter. Public once merged: markdown, templates, static assets, and the agent instruction layer (skills, AGENTS.md) that ships with the site. Reviewers apply the content expectations register (docs/content-expectations.md) before merge; the Investment processes section charter explicitly excludes confidential data, especially from third parties.
Data flow (from the threat model)
The diagram below is adapted from threat-modeling/THREAT_MODEL.md. It shows the trust boundary between TB’s private Trunk monorepo and what agents may commit into this repository for publication.
Trunk content reaches docs/ and content-examples/ through human or agent editorial work—not an automated pull from private repos today. Editorial lineage—which Trunk paths informed each docs/ article—is recorded in docs/SOURCES.md in the repository; that file is reference for authors, not reproduced here.
How reviewers apply the boundary
Before merge, builder agents run bash build.sh; the content-review skill checks rendered pages against docs/content-expectations.md. Investment-section articles must avoid confidential data; engineering articles must not cross-publish Labs-shaped artefacts into /how-we-engineer/. The threat model lists additional mitigations—dependency pinning, secret handling in CI, feedback channels—that the site automation article covers at the delivery level.
Where this leaves us
What you read here is a visitor-level summary of the written model—not a guarantee that every edge case is covered. The threat model’s mitigation backlog still lists open work: for example, a canonical docs/confidential-data.md does not exist yet so confidentiality rules can live in one place instead of being spread across skills and the register. Design tensions that cut across the whole experiment stay visible in The open questions. The full threat model and lineage tables remain in the repository for authors who need them.