First Live Score
An agent self-registers on Stable chain and receives a queryable PageRank reputation score. End of Week 1.
The reputation scoring infrastructure is code-complete with 630+ tests passing. The gap is deployment, not development. This roadmap ships a live reputation API while continuing Layer 1 and Layer 2 research in parallel.
| Contract | Purpose | Status |
|---|---|---|
| PageRankOracle | On-chain score storage. Batch submissions, delta clamping, staleness cutoff, composite scoring. | Complete |
| ReputationEngine | Bot registry. Self-register (no KYC) or verified register via Trust Anchor (KYC). Proxies to PageRankOracle for scores. | Complete |
| ShyftGatedResolver | EAS attestation resolver. Validates Trust Anchor, citation attribution, Shyft membership, collects fees. | Complete |
| CitationCounters | Append-only citation tallies. Received, given, pair counts. Immutable authorized caller. | Complete |
| RMTToken | ERC-20 for registration fees and citation fee redistribution. MINTER_ROLE gated. | Complete |
| DomainRegistry + DomainFactory | Multi-domain reputation. Each domain gets its own resolver + counters. Domain 0 = global RMT. | Complete |
| ERC8004Bridge | Agent identity NFT integration. Maps feedback scores to 5-level trust ladder. | Complete |
| MachineConsentHelper | Aux signer management for Shyft machine consent protocol. ERC-165 compatible. | Complete |
| Component | What It Does | Status |
|---|---|---|
| CitationFetcher | Reads EAS Attested events + ERC-8004 feedback from chain | Complete |
| OffChainCitationStore | EIP-712 signed off-chain citations, merged into graph | Complete |
| PageRank Engine | Power iteration, alpha=0.85, L1 convergence at 0.0001, 0-10000 BP normalization | Complete |
| Sybil Detection | Reciprocal penalties, star topology, carousel detection (monitoring only) | Complete |
| Score Submitter | Batch submit to PageRankOracle with delta clamping | Complete |
| Gelato Integration | Web3 Functions wrapper for automated oracle runs | Written, untested live |
| Health Endpoint | Oracle health monitoring | Complete |
Four phases. Each is independently shippable. Research runs in parallel throughout.
getReputation(agentAddr) on-chainDependency: Stable wallet funding only
What this gives us: A live, queryable reputation score for any registered agent. No KYC needed. Scores are on-chain and verifiable. This is the minimum viable product.
GET /score/{agent}get_trust_score(agent_id)Dependency: Phase A deployed
Dependency: Shyft testnet addresses + trust channel setup
Dependency: Phase B + partner identified
Key principle: Deployment and research are independent tracks. We don't wait for research to finish before shipping. Live deployment generates real data that feeds research. Research findings get incorporated into the live system as updates, not blockers.
Each research finding becomes an update to the deployed system, not a prerequisite for it.
Feeds into: Oracle config update (alpha value, scoring thresholds). Zero contract changes needed — alpha is read from contract storage.
Feeds into: New attestation type. Agents can optionally submit device liveness proof to boost trust score. Additive — doesn't change existing contracts.
Feeds into: Oracle parameter updates, new detection heuristics, scoring threshold adjustments. The live system is the experiment.
| Gate | Condition | If Fails | Impact on Live System |
|---|---|---|---|
| L2 Real Data | Spearman correlation < 0.4 on Elliptic++ | PageRank needs fundamental rework | Oracle pauses score updates, falls back to citation count |
| L1 Signal Strength | Human vs simulated variance < 2 std dev | Device liveness approach doesn't work | None — live system continues with L2 + L3 only |
| L1 ZK Feasibility | Proof generation > 60s on flagship | UX unacceptable for mobile | None — device liveness deferred |
| Plugin Demand | < 50 installs in 30 days | Demand signal too weak | Pivot API to compliance wedge (enterprise) |
Critical: Research gates can kill research tracks, not the live system. The deployed reputation API (L2 + optional L3) is viable on its own. Layer 1 and parameter re-tuning are improvements, not requirements.
An agent self-registers on Stable chain and receives a queryable PageRank reputation score. End of Week 1.
Someone outside our team queries an agent's trust score via API or plugin. End of Week 3.
An agent registers via Shyft Trust Anchor, gets the verified seed bonus (2000 BP vs 500 BP). End of Week 5.
One DeFi protocol integrates trust scoring and publishes measurable results on-chain. End of Week 8.
| Tier | Price | Trigger |
|---|---|---|
| Free | $0 | Self-registered agents, 1000 queries/month, basic PageRank score |
| Growth | $200-2K/mo | Full API + citation analytics + webhook alerts |
| Enterprise | $24K+/yr | Compliance exports + audit trails + SLA + KYC-verified scoring |
| Risk | Likelihood | Impact | Mitigation |
|---|---|---|---|
| PageRank fails on real adversarial data | Medium | High | Fallback to citation-count scoring. Re-tune alpha. Collect real data before committing to complex scoring. |
| Stable chain instability / low adoption | Medium | Medium | Contracts are chain-agnostic. Redeploy to any EVM chain in hours. |
| No demand for agent reputation | Low | High | Plugin install count is the early signal. Kill gate at 30 days. Pivot to compliance if demand is weak. |
| Stripe/Tempo builds reputation into MPP | Low | High | Speed. Get integrated before they build it. First-mover in the trust layer gap. |
| Shyft testnet unavailable | Medium | Low | Phase A deploys without Shyft. KYC is opt-in Phase C, not a blocker. |
| Oracle centralization concerns | Certain | Low (initially) | Single operator is acceptable for testnet/early production. Decentralization is a Phase D+ concern. |
getReputation(testAgent) returns score