Anchor paper · Working paper v1.0
Sovereign Knowledge Federation
A boundary-and-capability model for sharing second brains between people and their AIs, without surrendering them to a platform.
The question it answers
Two people each keep a careful, structured record of what they know, and each has an AI that can reason over it. One of them has already worked out something the other needs. Today there is no good way to move that knowledge across. It can be published flat to the open web, pasted into a chat, or handed wholesale to a platform that then owns it, and none of those lets the second person's AI treat the first person's knowledge as knowledge while still knowing whose it is and how far to trust it.
That missing option is the subject of the paper. Personal knowledge management matured into second brains; AI matured into assistants that can read them. Federation between sovereign vaults is the layer nobody has built, and the paper develops it from first principles, attacking each layer for its failure modes before proposing a resolution.
The shape of the argument
Structure, access control, and federation turn out to be three readings of one primitive: a securable boundary, cleanly separated from the capabilities that grant the right to cross it. Knowledge that crosses a boundary never travels as a bare fact; it travels wrapped in provenance, who asserted it, how they came to believe it, and how many hands it passed through. Trust is computed by the receiver at the moment of use, never granted by the network, because a system that refuses to adjudicate truth is precisely the one that cannot be captured as a truth authority.
A few of the working terms, because the vocabulary is half the fun:
One person's or organization's entire sovereign knowledge sphere. The unit that federates.
A knowledge bridge between Universes: a capability whose grantee lives outside your boundary.
A foreign claim in transit: observed and weighed while it passes, never silently adopted as your own.
Now published as a v1.0 working paper. It is released to establish authorship and to invite collaboration, with its priority recorded by a signed commit and an OpenTimestamps proof, the same provenance discipline the paper argues for. Read it above, or read the source and open issues on GitHub. Comments and attacks are welcome.