Orogen
Open Playground

Roadmap

Shipped, planned, and what comes next.

The W0 and W1 build waves are shipped as code and verification artifacts. The Forge testnet is live, and the multi-firm chain audit comes before mainnet. The rest of the arc, from permissionless onboarding and the ETH bridge through mobile inference and federated post-training, is sequenced so each phase depends on real results from the one before it. Each milestone is tied to measurable conditions, so if a condition takes longer, the next phase waits for it.

Plate 02 · roadmap arc

W0 → W1 → Forge testnet → audit → mainnet TGE → permissionless

  • shipped build wave
  • Forge planned
  • audit (planned)
  • mainnet TGE (future)
  • permissionless (future)

W0 and W1 are shipped as implementation work. The Forge testnet, the multi-firm audit, mainnet, and permissionless onboarding follow. Each milestone has measurable conditions, so if a condition takes longer the next milestone waits for it. The conditions for each node are detailed in the phase table below.

Phase detail

Targets, gates, KPIs.

The first four phases map to the visual arc above (two shipped locally, Forge planned, audit planned). The remaining five are sequenced after mainnet TGE: permissionless onboarding, mobile + bridge, RL pilots, federated post-training, and the Phase 4 pretraining-from-scratch go/no-go.

  1. SHIPPED Shipped (W0) Phase 1 / 9

    Stealth dev

    Target

    Centralised inference stack stood up (vLLM V1 + LiteLLM gateway). Operator-partner integration path designed against H100/H200 CC mode. First end-to-end signed-receipt path through the gateway.

    What it depends on

    Stack feasibility validated: gate cleared.

    KPIs
    • Gateway routes OpenAI-compatible traffic end-to-end
    • H100/H200 CC mode attested in operator-partner lab integration
    • p99 latency < 800ms for 8K-prompt / 1K-output on Llama-3.1-70B in lab
    • TEE attestation overhead < 5% in lab measurement
  2. SHIPPED Shipped (W1) Phase 2 / 9

    Permissioned operator partners

    Target

    Pallet suite (pallet-bme, pallet-model-registry, pallet-operator-stake, pallet-job-market, pallet-yuma-consensus, pallet-slashing) written and integrated. L1–L4 verification stack composed end-to-end against the operator-partner pool. AMD MI3xx integration started.

    What it depends on

    Verification-stack composition validated end-to-end: gate cleared.

    KPIs
    • Pallet suite v0.9 integrated against the chain-node fork
    • Verification mismatch rate < 0.1% of sampled requests in pre-testnet runs
    • Operator-partner pool wired against the L1–L4 stack
    • Adapter publishing path running end-to-end against pallet-model-registry
  3. FUTURE Preparing (Forge) Phase 3 / 9

    Forge testnet

    Target

    Forge public testnet is the next integration target. Seed validator, chain-indexer + GraphQL, OpenAI-compatible traffic path, and attested operator onboarding must be brought up before this phase is called live. Target second-vendor profile: AMD MI3xx and Blackwell B200 CC operators alongside H100/H200. Pallet suite covers pallet-bme, pallet-model-registry, pallet-operator-stake, pallet-job-market, pallet-yuma-consensus, pallet-slashing (pre-audit).

    What it depends on

    Mainnet depends on second-vendor (AMD) operators reaching production grade alongside NVIDIA. If that takes longer, mainnet moves with it.

    KPIs
    • 15+ operators across NVIDIA + AMD on testnet
    • Verification mismatch rate < 0.05% over sampled traffic
    • p99 latency < 500ms on testnet routes
    • 100+ adapters published via pallet-model-registry
    • Day-N replay coverage ≥ target sample-rate floor
  4. FUTURE Planned (audit) Phase 4 / 9

    Multi-firm chain audit

    Target

    Three independent firms will review the chain pallets, the gateway burn path, and the slashing schedule in parallel after Forge is operating. Slashing pallet stress tests and CRL drills must pass before this phase is marked complete.

    What it depends on

    Mainnet is conditional on the audit closing clean. Open findings reset the clock until they are resolved.

    KPIs
    • Three independent audit reports filed
    • All critical and high findings resolved on Forge
    • Slashing pallet stress test passing under adversarial conditions
    • CRL drill rehearsed end-to-end on Forge
  5. FUTURE After audit Phase 5 / 9

    Mainnet TGE

    Target

    Token live on Substrate fork mainnet. BME loop active from Day 1, with real burn, real mint, and every customer transaction settling on-chain. All Forge operators and customers migrate to on-chain mainnet settlement. EVM JSON-RPC layer live; ETH bridge live.

    What it depends on

    If Day-30 burn < 0.3× mint, the BME loop is not capturing customer demand. Pause permissionless onboarding; investigate gateway-burn integrity.

    KPIs
    • TGE clean (no chain halts in first 30 days)
    • Day-30 cumulative burn > 0.5× Day-30 cumulative mint
    • Token listing on ≥ 2 mid-tier CEXes within 30 days
    • Day-90 subsidy ratio < 2×
    • Operator count growing (permissioned-only is acceptable)
  6. FUTURE Post-TGE Phase 6 / 9

    Permissionless onboarding · LoRA marketplace · opML layer

    Target

    Permissionless operator onboarding. Permissionless adapter publishing. opML challenge window deployed (Layer 5). Cross-region routing matures. Permissionless validator entry; Yuma slashing live.

    What it depends on

    If verification mismatch rate climbs above 0.1% after permissionless opening, tighten operator registration requirements (raise stake floor at prosumer/edge).

    KPIs
    • 30+ operators (permissionless)
    • 200+ adapters in catalog
    • 50+ customers paying
    • $200K MRR
    • p99 latency < 400ms
    • Verification mismatch rate < 0.02%
    • Yuma consensus running with ≥ 64 validators
  7. FUTURE Post-TGE +2 quarters Phase 7 / 9

    Mobile inference · bridge launch · RL preparation

    Target

    Mobile-tier inference (Apple-silicon edge tier) productionised behind opt-in zkML for embeddings and classification. ETH ↔ Orogen bridge expanded with Snowbridge or equivalent. RL marketplace specification frozen (Phase 2 of training roadmap). Atropos task definitions accepted on testnet.

    What it depends on

    If no measurable RL customer demand by gate-review, postpone Phase 3 (federated post-training) and double-down on inference + LoRA.

    KPIs
    • $400K MRR (inference + LoRA + RL pilots)
    • ≥ 1 customer-paid RL training run completed on testnet
    • Operator count > 50
    • Subsidy ratio < 1.5×
  8. FUTURE Post-TGE +4 quarters Phase 8 / 9

    RL pilots at scale

    Target

    Multiple concurrent INTELLECT-2-pattern RL runs (32B-class with math/code verifiers). TOPLOC-style activation attestation for rollouts integrated. First externally-customer-paid post-training run.

    What it depends on

    Phase 3 (federated post-training) enters production only if sustained $200K+ MRR is attributable to training at this gate.

    KPIs
    • $700K+ MRR
    • Operator count > 70
    • Subsidy ratio < 1.2×
    • First decentralised post-training cohort (7B-class) bolt-on operational
  9. FUTURE Post-TGE +6 quarters Phase 9 / 9

    Federated post-training

    Target

    DisTrO / SparseLoCo bolt-on for 7B-class decentralised post-training in production. $1M/mo MRR. Decision: enter Phase 4 (pretraining-from-scratch market)?

    What it depends on

    Phase 4 entry is conditional on (a) sustained $12M+ annualised revenue and (b) subsidy ratio < 1×.

    KPIs
    • $12M+ annualised revenue
    • 100+ operators
    • Subsidy ratio < 1×
    • Phase 4 go/no-go decision rendered

Cross-phase invariants

What does not move.

  • The BME loop runs unbroken from TGE Day 1; there is no "pay USDC and skip the token" backdoor at any phase.
  • The verification stack is L1–L8 throughout; new layers may be added (e.g. ZKML for embeddings), but no layer is dropped.
  • The 75 / 15 / 5 / 5 settlement split is governance-adjustable at launch but constitutionally bounded, and the provider share cannot fall below 60%.
  • The 5% supply/year mint ceiling is hard. The chain rejects mints above it. Governance can lower it; raising it requires constitutional amendment (supermajority, 60-day timelock).
  • Multi-vendor TEE at launch. Single-vendor monoculture is treated as a security regression and reverts to permissioned-only operator onboarding until restored.
Tokenomics detail Verification stack FAQ