Webπ · a coinage Designed
The value your information creates should return to you.
That is the whole argument, and Webπ is the name it goes under here. Web3 set out to return ownership of digital assets to the people who hold them; this asks for the same over what a person actually produces all day — memory, creation, and the context of their work.
It is not a specification. There is no Webπ RFC, no working group, no wire format and no registry. Nobody outside this project uses the term. It names a position, and the position has to be argued for on its merits rather than borrowed from the authority of a version number.
What changes
Three shifts, stated so they can be disagreed with.
The project ledger
Eight pieces of work, and the two that already run.
Not one token. Two of the eight exist as running code; the other six are documents.
- 01 centram-osDistribution A bootable operating system built on Linux, for using a personal AI node as an everyday machine. Nothing is bootable yet. The OS page. Designed
- 02 centram-corePermission boundary Identity, permission, consent and audit. Exists as a running daemon with a test suite. Implemented
- 03 personal-intelligenceThe assistant Continuous support for a person’s intent, memory and work. A research direction, not a component. Designed
- 04 modulesCapability providers Documents, search, translation and development, each executing only inside a scope Core issued. Specified, not built. Designed
- 05 riverTyped data plane Carries local events, synchronisation and shared verification results. It grants no authority of its own. A substrate baseline is implemented. Implemented
- 06 intelligence-assetsOwned data assets Assets born of memory and capability, carrying their provenance and their terms of use. Prototyped locally; no conformance gate has been passed. Designed
- 07 cognitive-graphCollective learning Verifying how capability grows across a network without sharing anyone’s raw data. The measurement so far is against it — see below. Unproven
- 08 centinelRights settlement Settlement after rights are established. Deliberately separate from Core’s permission model. Not implemented. The economy page. Designed
The hinge Measured negative
One result decides whether any of this is worth building.
Every collective claim on this page rests on one assumption: that pooling what separate machines learn beats both the best specialist and a generalist trained on everything.
That was run. At 1.1B parameters, over five seeds, on a held-out task,
it did not hold: loss_task_bridge_holds = false, the verdict
was lost_to_single, and the margin was
−0.0804. The win in loss space was real and
reproducible; the crossing from there to task performance was not.
The project’s own audit backlog treats that as the single go/no-go for all network and token work, and says plainly that if the bridge never closes, the collective superstructure adds no product value. A larger run is the open question, not a formality.
What survives either way is the part that never depended on the network: a local, token-free personal AI that cannot be made worse by anything it receives. That is the floor, and it is a real floor.
Working on it
Information is your creation.
The argument needs people to attack it as much as people to build it.