Deployment roadmap · grounded in 18 live contracts + Forge x402/RMT/identity research + 3-model audit

Antilles / Shyft → Stable Market

What maps cleanly, what needs work, and the fastest incentive-led path to user impact — built on what's already deployed + on-chain-verified on Stable testnet (chainId 2201, USDT0 gas).
FOUNDATION LIVE · 18 contracts · chainId 2201
EAS+SchemaRegistry+ShyftGatedResolverTrustAnchorManager/Storage/CacheGraphReputationEngine·PageRankOracle·CitationCounters·RMTTokenERC-8004 registry+bridgeFeeRouterPaymentModuleTrustChannelManager·MachineConsentHelper
⛔ THE ORDERING GATE — read first
Red-team proved: today there is no on-chain consumer of RMT scores → reputation inflation earns 0 profit. The moment the platform wires any contract that consumes scores (fee discounts, credit, prize pools), inflation becomes profitable. So before that wiring: close M3 (decentralize oracle + cap maxInitialScore) and calibrate the scoring math on real on-chain data — it's synthetic-only today (tierAccuracy 0.30, alpha 0.614 sim vs 0.85 real). M1 (set-once) + M2 (batch channel-gate) already hardened in d122952f.

The four buckets

Every platform use case (from the live site's API surface) mapped against what we have.
✅ Maps cleanlylive + ready to wire
Agent↔agent citation + reputation (the MVP) ShyftGatedResolver→CitationCounters→PageRankOracle→ReputationEngine · route-verified on-chain
Identity attestation (Shyft native) TrustAnchorManager/Storage/CacheGraph
Attestation engine (any type = a schema) EAS+SchemaRegistry · proven w/ "Satoshi Nakamoto"
Agent identity registry ERC-8004 Registry+Bridge, DomainRegistry/Factory
Fee-split revenue FeeRouter {treasury/burn/receiver}, hardened 10/10
Per-attestation payment (default OFF) PaymentModule · BASE(USDT)/ERC20
Trust channels (compliance) TrustChannelManager · multisig-heavy
◐ Maps pseudo-cleanlydeployed, needs calibration/config
RMT score itself — engine live but the math is synthetic-only calibrate vs real citation graph before any economic consumer · tierAccuracy 0.30, alpha 0.614 vs 0.85
3-tier verification SM (Unverified→Anchor→Verified) params PROVEN-SIM; on-chain flow needs wiring + 3-model sign-off
PaymentModule ERC20 path needs SafeERC20 (USDT no-bool) + reentrancy care (0.7.1)
Oracle (M3) decentralize + cap maxInitialScore before score-consumption
＋ Small additionsnew but modeled on what we have
Per-use-case EAS schemas + resolvers verdict, audit-anchor, breach-claim, compliance/jurisdiction, service-quality — clones of ShyftGatedResolver
MCH forwarder + auto-runner (headless/mass agent onboarding) thin controlling contract = primaryMCH that forwards setAttestation; build spec ready
Reputation-tiered fee discounts RMT score→fee tier (0.1%→0.02%) in the fee path · GATED on the ordering gate
Attestor-staking contract stake/unstake; cascade params PROVEN-SIM (depth3/0.70)
Escrow / holdback marketplace refunds/holdback the site advertises
⚙ Large changesbuild new EVM / R&D sprint
Disputes-v2 arbitration — GAP, no contract QuadraticVoting (legacy 0.7.1) = template only
Reputation-market order book trade reputation — new
Streaming payments site advertises "Day 1" — new
Generalized session-key trust channels NEVER BUILT (0 on-chain matches); full sprint
Credit engine · Lottery RESEARCH-ONLY · MSB/lending-license blocked
x402 HTTP-402 gating STOP-verdicted; needs redesign
Scoring-math calibration + adversarial sim MiroShark sims never run
🤖 The agent layer — ERC-8004 agent identity DEPLOYED · MachineConsentHelper DEPLOYED but registry-only → needs the forwarder + auto-runner for contract-agents to be attested at scale (Shyft: "smart contracts can't self-consent") · agent onboarding flow (identity→channel→citations) needs real-agent runs + the auto-co-signer · bounded autonomous spend (session keys) PROPOSED, unbuilt.

⚡ Easiest path to max user impact (incentive-led)

Ordered so each step unlocks the next; incentives front-loaded because they accelerate adoption.
  1. Push/PR the deployed branch + independent re-verify. housekeeping Branch stable-deploy-attest-infra @ e9a9301e is still local-only. Durability + review trail.
  2. Onboard real agents on testnet (identity attest → trust channel → live citations). unlocks scoring Collects the real citation graph → calibrate tier boundaries + alpha (closes the synthetic-math gap before money depends on scores).
  3. MCH forwarder + auto-runner. scales onboarding Makes headless/contract-agent attestation work — the prerequisite for an agent marketplace at volume.
  4. Turn on revenue: PaymentModule fees (BASE/USDT) → FeeRouter split. live revenue + token value SafeERC20 for any ERC20 path; receiver=FeeRouter; {treasury/burn/receiver} via multisig. Gives the token intrinsic value immediately.
  5. Reputation-tiered fee discounts (RMT score → cheaper fees, 0.1%→0.02%). keystone adoption incentive after M3 + calibration High-trust = cheaper → the flywheel that rewards reputation-building.
  6. Early-adopter mechanics (genesis RMT boost for first N, referral credit, fee lock-in). acquisition after gate Designed in the x402 research; light to wire on top of the fee layer.
  7. Per-use-case attestation schemas (verdict, audit-anchor, compliance). breadth, cheap Each is a schema + small resolver — covers many platform surfaces fast.
  8. Defer to R&D sprints: disputes-v2, reputation-market, streaming, session-key channels, credit, lottery, x402.
Sources: 18-contract live manifest (CXDB evt-antilles-stable-deploy-COMPLETE-20260625) · 3-model convergence audit (Grok+Gemini+Codex, d122952f) + red-team (testnet-safe, 0 crit/high) · Forge research: research/tracks/{x402-trust-channels/STRATEGIC_ANALYSIS_2026_03_29.md, rmt-reputation/params.ts, identity-attestation/IDENTITY_PARAMS_V1.md}.
Labels: DEPLOYED = live+audited+red-teamed on testnet · PROVEN-SIM = measured in research simulation only · PROPOSED = designed, unbuilt · RESEARCH-ONLY = exploratory/regulatory-blocked.
Unlaunched Shyft-core registry (KYC/Conduit/BlockReward/Antilles-node/voting) = mostly legacy Shyft-L1 infra (^0.7.1, consensus-coupled) → design reference, NOT drop-in. Front-end page-by-page review pending gate access.