cyber/launch

A self-verifying knowledge graph where attention, computation, and consensus converge into a single metric (φ*), enabling intelligence emergence without central control.

nox Optimizing civilization's ability to know what matters

Version: 2026.06 | Status: foundations complete → integration spine

How to read this plan

Three anchors fix the whole path:

• MVP testnet — the first network where two independent nodes agree on one cybergraph, every signal carrying its own proof.
• Mainnet — the launch that must be correct on the first try, because no patch relay exists between stars.
• The structural milestones between them — each one a capability the network gains, each one a hard gate that opens only when its test passes.

A milestone is code that passes its gate, never a date on a calendar. Durations below are work estimates in sessions, not deadlines. The order is a dependency order: a milestone starts when its predecessors pass.

Where we are — 2026.06

The stack was built bottom-up, so it inverts the textbook order. The hard cryptographic floor — field, hash, commitments, the proof-native VM, the proof system, authenticated state — is implemented and tested. The frontier is the integration spine above it: the consensus that lets nodes agree, the runtime that drives the loop, and the single binary that wires them into a node.

The bostrom network has run 3+ years as the bootloader — ~70K neurons, 2.9M cyberlinks, 3.1M particles — and is the migration source, not the soft3 network. The soft3-native node does not exist yet; building it is the spine of this plan.

Theoretical foundations established:

• Convergence guarantee: unique φ* exists, exponential convergence, bounded mixing time
• Conservation law: Σφ*ᵢ = 1, always — no inflation, no leakage
• GNN isomorphism: tri-kernel update ≡ multi-channel graph neural network message pass
• Transformer equivalence: CGC focus ≡ iterated sparse attention with economic grounding
• convergent computation: replaces the halting problem — the system converges, never halts
• Free energy minimization: Δφ* is literally the gradient of system free energy

The milestone ladder

M0 — Foundations (complete)

The floor every other milestone stands on. Implemented and tested today:

• strata — the five algebras (Goldilocks, F₂¹²⁸, R_q, tropical, isogeny) under one trait stack
• hemera — Poseidon2 over Goldilocks; particle identity, the nox hash jet. Audit pending before any address is frozen.
• lens — Brakedown, Binius, Ikat, Assayer, Porphyry, one commitment per algebra
• nox — the proof-native VM; 18 reduction patterns (16 compute + call + look) plus jets, every run a stark by construction
• zheng — SuperSpartan + Brakedown + sumcheck; prove and verify wired
• bbg — authenticated state, the mutator set, query proofs
• radio — QUIC transport with Poseidon2 verified streaming, gossip, and a docs store

Open inside this floor, carried as work into later milestones: hemera audit, zheng look-argument soundness, nox parallel reduction, bbg private record scanning.

The road to MVP testnet

Goal: two or more independent nodes converge on one cybergraph — signals gossip, order into per-neuron chains, apply their cyberlinks to state, and reach φ*ᵢ > τ finality, with every signal carrying a proof the neuron did what it declared.

In scope: signal lifecycle, proof at seal, gossip, minimal consensus, a node binary, the value layer ($CYB emission and rewards), and φ* ranking — an MVP with a live economy and focus, not a bare consensus net. Deferred past the testnet (see P-milestones): privacy circuits, recursive proofs, sharding, formal verification, and the chain migrations. The testnet proves the architecture agrees and the economy runs; it does not yet hide owners.

S1 — Wire & framing

tape is the typed frame every particle and signal travels in. The encoder/decoder is partial and the wire format is a draft.

Estimate: 2-3 sessions.

S2 — Identity

mudra gives a neuron its keys and signatures. The crate is a scaffold and its build is broken on path dependencies. The testnet needs signing and addressing only; FHE, stealth, and threshold are P-milestone work.

Estimate: 1-2 sessions for the testnet subset.

S3 — Proven processor (single node)

cybergraph runs local-first today: intend / seal / link apply cyberlinks to bbg, query runs inf. The missing property is the alignment guarantee — seal(i, s) accepted iff σ(s) ⊢ scope_hash(i). This closes the zheng look-argument soundness gap at the same time.

Estimate: 3-5 sessions.

S4 — Networking

radio moves bytes; foculus orders them. Today foculus is pull-only — peers fetch on demand, signals do not propagate. A network needs a mempool membrane that pushes a new signal to peers.

Estimate: 3-4 sessions.

S5 — Consensus v0

foculus is a complete specification with zero code — the largest single gap. A particle is final when φᵢ > τ. The testnet needs the minimal core: compute φ over committed state, apply the fork-choice rule, declare finality.

Estimate: 6-10 sessions. This is the critical path.

S6 — Node + genesis

No binary wires the spine together, and there is no genesis. soma is the runtime that drives the fetch → execute → prove → commit loop; the testnet needs its minimal drive-loop, not the full cognitive architecture.

Estimate: 5-8 sessions.

S7 — Economics

tok carries value: coin (fungible) and card (non-fungible), all change a plumb mutation under conservation. Rewards mint on Δφ* — creating valuable structure is creating value — Sybil-resistant via stake-weighting. The testnet runs a live $CYB economy from day one, so price and rewards are exercised, not deferred.

Onboarding is rewarded by a referral built into the protocol: a referrer earns a lifetime cut of the stake-weighted Δφ* their referees generate — never a per-head bounty. Paying on focus created, not headcount, keeps the referral Sybil-resistant (a fake account creates no focus, so referring it pays nothing) and turns early movers into growth agents aligned with the network rather than dumpers.

Gate: conservation enforced by proof; reward and referral equilibrium simulation-tested under 100× adversarial load, including Sybil onboarding.

S8 — Ranking

tru computes the field today; the CT-0 model-compilation passes bail. This milestone runs the tri-kernel over the network's graph, deriving models from φ*, adversarially proven — so the testnet ranks by focus, the metric everything else serves.

Gate: φ* over the network graph, explicit Lyapunov function with dV/dt < 0, bounded locality k = O(log(1/ε)).

S9 — Client (cyb)

The sharp emission head is justified only if the network is useful from block one — a front-loaded incentive converts no one onto a thing that does not yet work. cyb is that utility: the browser and avatar where a neuron writes cyberlinks, sees focus, and queries the graph. It exists and builds today; this milestone wires it to the soft3 node so value lands immediately, not after years of protocol maturation — the gap that left earlier chains' tokens speculative until utility caught up.

Gate: a neuron installs cyb, creates a cyberlink, sees it ranked by φ*, and queries the graph — end to end, day one.

MVP testnet — the gate

The testnet is the proof that the architecture converges with a live economy: the dumb processor (cybergraph) and the smart runtime (soma) drive one network to agreement, carrying $CYB and ranking by φ*. Owners are still public and the bootloader graph has not migrated — that is the work from here to mainnet. The first move past the gate is the $PUSSY migration rehearsal (R1).

Structural milestones: testnet → mainnet

Each one a capability the testnet lacks, ordered by dependency. ($PUSSY migrates first, right after the gate — see Migration.)

P1 — Privacy circuits

UTXO-style privacy with ZK proofs for every state transition. Public: edge existence, aggregate energy per particle, the φ* distribution. Private: neuron identity behind an edge, individual energy ownership, link authorship. φ* stays computable from public aggregates alone — secure multi-party computation of a GNN forward pass. Depends on bbg private scanning and mudra stealth.

Gate: transaction + cyberlink circuits, nullifier soundness, leakage budget L(queries, graph) bounded.

P2 — Recursive proof

The stark verifier is itself a nox program: proofs can be verified, and verification can be proven. Unlocks O(log n) light clients.

Gate: inner verification circuit arithmetized, light-client verification of any state claim.

P3 — Sharding & data availability

Shards as subtopoi, a sheaf of attention weights keeping cross-shard focus consistent. foculus has erasure coding and DAS commitments; the sample verifier is missing.

Gate: DAS verifier complete, cross-shard consistency proven, gossip bandwidth ∝ stake.

P4 — Formal verification spine

Runs parallel to every milestone, converging here. Each line below is one of the Pre-Launch gates.

Migration — the bootloader chains come home

The soft3 network launches with state, not empty. Two live Cosmos chains carry the graph and its neurons today; the migration brings that graph — particles, cyberlinks, neurons, karma — home onto the soft3 network, lowest stakes first. $PUSSY migrates right after the MVP testnet as the rehearsal; $BOOT and the crystal come at mainnet. The $CYB genesis is separate from all of this: it is the $C-holder snapshot (≈ 1% of supply); $BOOT and $PUSSY balances do not convert.

R1 — migrate $PUSSY (the rehearsal)

space-pussy is the experimental net; migrating its graph first proves the whole pipeline — export state, map it into cybergraph form, verify bijectively — without risking bostrom. A dress rehearsal at low stakes, run as soon as the testnet stands.

Gate: full space-pussy graph (particles, cyberlinks, neurons, karma) reproduced on the soft3 net with zero loss; the replayed root matches.

R2 — migrate $BOOT (the real one)

The same pipeline at mainnet scale: every neuron, cyberlink, and particle from bostrom — ~70K neurons, 3.1M particles, 2.9M cyberlinks, three years of history — with karma carried as reputation. The graph comes home; the separate $C-holder snapshot taken here is what mints the genesis 1% ($BOOT, $H, $V, $A, and $PUSSY balances do not convert to $CYB).

Gate: bijective state mapping, zero data loss; frozen snapshot signed; karma reconciles; every particle re-addressable under hemera identity.

R3 — align the genesis crystal

The migrated graph is organic and large; the crystal is the curated 5,040-particle irreducible seed — the grammar and vocabulary. Alignment makes the crystal the foundational lattice and the migrated knowledge the flesh hung on it, deduplicated against the seed. The 12 genesis invariants are the gate.

Gate: 12 invariants green (completeness, connectivity, reachability ≤ 6 hops, irreducibility, positivity, …); the crystal is the basis; zero orphans, zero red links.

Mainnet genesis is then this migrated, crystal-aligned state going live under the soft3 protocol and the $CYB emission curve.

Mainnet — the gate

No patch relay exists between stars. What launches must be correct. Before launch, answer six questions — five with machine-checked evidence, one with the record of real use:

All six green → launch. Any red → no launch. No exceptions.

The light-cone is merciless. What you ship is what arrives.

Token Architecture

Four Token Types (Protocol-Native)

$CYB is the consensus token of the full cyber network. On bostrom (bootloader): $BOOT (stake/fees), $H (liquid fuel), $V (will), $A (attention).

Adaptive Economics

Three PID-controlled variables adapt automatically — no governance vote needed for routine adjustment:

α (allocation curve exponent): balances PoW vs PoS allocation. staking_share = S^α.

φ (security floor): minimum issuance for security. φ ≥ k · (TVL/MarketCap) · r.

β (fee burn rate): decouples gross rewards from net inflation.

Staking yield at equilibrium: r_s = (G · S^(α-1)) / M. Master safety indicator: ρ = d(Attack Cost)/dt ÷ d(Attack Profit)/dt; ρ > 1 means defenses grow faster than threats.

Supply and emission

Total supply is the Goldilocks field order (p = 2⁶⁴ − 2³² + 1). Genesis seeds a 187,416,084,623,451,570 $CYB starting balance for $C holders (≈ 1% of total); the rest emits along a power-law curve — a hot bootstrap head and a heavy multi-decade tail. Full supply cap, genesis, and the emission curve live on $CYB.

The genesis seed

The crystal is the genesis content — exactly 5,040 particles forming the irreducible basis from which all civilizational reasoning composes. An alphabet of a mind.

Two-layer load: a lattice (4,392 particles, ~454K tokens) of structural vocabulary that fits one model context, and flesh (648 particles, ~1,165K tokens) of articles and proofs retrieved on demand via cyberlink traversal. The 12 genesis invariants (completeness, connectivity, reachability ≤ 6 hops, irreducibility, positivity, self-reference, bridge density, type balance, defect freedom, growth-readiness, narrative depth, self-explanation) are the quality gates before the crystal ships. See crystal for the full specification.

The endgame

A living, self-optimizing knowledge network that:

1. Learns from every form of input on Earth — humans, AI, sensors, biology
2. Holds coherence under extreme conditions, including interplanetary latency
3. Evolves without central authority — governance through focus dynamics and futarchy
4. Maximizes the survival, intelligence, and flourishing of the planet's biosphere
5. Proves every claim — no trust required, only math

The network IS thinking. No node comprehends. The network knows.

Cross-references

• crystal — the genesis seed specification
• cyber/tokenomics — the 7-mechanism incentive spec
• learning incentives — reward design, link valuation, attribution
• cft — the collective focus theorem
• soft3 — the stack these milestones build on
• cybergraph and soma — the dumb processor and the mind that drives it