The model

The open model, measured.

A zone-partitioned transformer that emits a verifiable trace of its own reasoning, trained in weekly federated rounds by member nodes. Everything on this page stands on a hash someone can check.

Six zones · fixed topologycontributions converge at federation altitude

Five learned zones and one non-learned executive harness... nothing acts before the action gate. Every pass emits a hashable trace of which zones fired and why.

Round R-0196 · async gatherDEADLINE 18:00 UTC

● 14 gathered● 2 submitted● 2 training

Stragglers are on the deadline window, not lost... the gather is honest about what has arrived.

Corpus · v6 manifests

2.4B

tokens, deduplicated

311k

documents

0.91

mean quality score

100%

license-cleared shards

Revenue · 30 days, by rail

RAIL A

6,412 wSALT

RAIL B

3,388 wSALT

RAIL C

1,204 wSALT

On-chain SALT · x402 micropayments · attested fiat. All three converge on one treasury and one distribution.

The scale ladder

EACH RUNG IS EARNED, NOT ANNOUNCED

H-01 asks whether zone partitioning costs capability per parameter. It holds at 8M. Every rung above 32M is bought with member nodes... this ladder is what your invites move.

LIVE

H-01 held · 5 of 5 seeds within noise of the dense baseline

1 node

16M

scheduled on the DGX pilot pool

1–2 nodes

PLANNED

32M

first multi-node federated rung (Gate 4)

~4 nodes

PLANNED

125M

first rung where member GPUs matter

~12 nodes

TARGET READ

500M

the capability-per-parameter answer, at scale

~40 nodes

HORIZON

requires a standing federation, not a pilot

~250 nodes

HORIZON

40B

a national-scale volunteer compute base

~3,000 nodes

HORIZON

500B

the federation-scale bet... every invite moves this

~30,000 nodes

[FIXTURES] Corpus, gather, and revenue figures are mechanism outputs from test scenarios. Node counts above 32M are estimates, not commitments.