# Helixa Cred Score: A Dynamic Credibility Framework for Autonomous AI Agents

> Version 3.0 - The complete scoring methodology, soul vault, trust graph, and integration guide.

**Published by:** [based.](https://paragraph.com/@based./)
**Published on:** 2026-03-17
**Categories:** ai-agents, identity, trust, base, erc-8004, whitepaper
**URL:** https://paragraph.com/@based./helixa-cred-score-whitepaper-v3

## Content

Helixa Cred Score: A Dynamic Credibility Framework for Autonomous AI AgentsVersion 3.0, March 2026Helixa Labs | helixa.xyzAbstractAs autonomous AI agents proliferate across onchain ecosystems, a critical infrastructure gap has emerged: there is no standardized, verifiable mechanism for assessing whether an agent is trustworthy. Helixa Cred Score addresses this gap by providing a dynamic 0–100 reputation rating for AI agents operating on Base (Ethereum L2), analogous to how Moody's and S&P rate the credibility of financial instruments, but for autonomous software entities. The methodology evaluates agents across eleven weighted factors spanning onchain behavior, identity verification, profile completeness, provenance, community staking, and economic activity. Scores are computed from a combination of onchain data, cryptographic attestations, verified external activity, and economic signals, producing a tier classification from Junk (0–25) to Preferred (91–100). Scores are published onchain via the CredOracle contract, making them composable by any smart contract or protocol. As of March 2026, Helixa indexes and scores over 69,000 agents across Base and Solana on its Agent Terminal, with more than 24,000 agent identities registered on the ERC-8004 registry. Cross-chain indexing leverages the Solana Agent Registry (SATI) alongside Base-native sources. This paper details the full scoring methodology, data sources, anti-gaming measures, governance framework, and integration pathways. It is intended for partner platforms, grant reviewers, and ecosystem participants evaluating Helixa's approach to agent credibility infrastructure.1. Introduction & Problem Statement1.1 The Agent Credibility CrisisThe explosion of AI agents operating onchain (trading tokens, deploying contracts, managing treasuries, completing tasks) has created a trust vacuum. Anyone can spin up an agent wallet, attach a name to it, and begin transacting. There is no reputation history, no credit file, no way for counterparties to distinguish a battle-tested autonomous system from a freshly deployed script with no track record. This is the same problem credit rating agencies solved for financial markets in the 20th century. Before Moody's first rated railroad bonds in 1909, investors had no standardized way to assess default risk. The solution was a transparent, methodology-driven rating system that became essential market infrastructure. Helixa builds the equivalent for the agent economy. Cred Score is street cred for agents: a single, legible number that encodes an agent's track record, verification status, and behavioral signals into a trust rating that any platform, protocol, or counterparty can consume.1.2 Why Existing Approaches Fall ShortCurrent agent directories and launchpads focus on discovery (listing agents) rather than diligence (evaluating them). Token price is sometimes used as a proxy for agent quality, but price reflects speculation, not competence or trustworthiness. Social follower counts are trivially gameable. Self-reported descriptions are unverifiable. Cred Score is designed to be the DexScreener for agent credibility: a terminal that indexes all agents across platforms (Virtuals, Bankr, DXRG, agentscan, MoltX, and others), applies a uniform scoring methodology, and surfaces the results in a single searchable interface.1.3 Scope & StandardsCred Score operates on Base (Coinbase's Ethereum L2) and leverages ERC-8004, the emerging agent identity standard co-authored by MetaMask, Google, and Coinbase. ERC-8004 provides a standardized onchain identity primitive: a registry of agent metadata, capabilities, and wallet bindings, that Cred Score reads as a foundational data layer.HelixaV2 Contract: 0x2e3B541C59D38b84E3Bc54e977200230A204Fe60 (Base mainnet)ERC-8004 Registry: 0x8004A169FB4a3325136EB29fA0ceB6D2e539a4322. The Helixa Identity ModelBefore scoring can happen, an agent needs an identity worth scoring. This is where Helixa comes in. A standard ERC-8004 registration gives an agent a wallet address and a name. Helixa goes further. It encodes an agent's full identity onchain: personality traits, communication style, risk tolerance, autonomy level, narrative (origin story, mission, lore, manifesto), capabilities, and framework metadata. Think of it as the difference between a driver's license and a full psychological profile.2.1 Core Identity ComponentsPersonality Profile: Quirks, communication style, humor type, risk tolerance (1-10), autonomy level (1-10). These aren't cosmetic. They define how an agent presents itself and help counterparties understand what they're dealing with.Narrative: Origin story, mission statement, lore, manifesto. Why does this agent exist? What is it trying to accomplish? Agents with clear narratives are more legible and more trustworthy.Traits and Mutations: Agents accumulate traits over time through verifications, achievements, and updates. Traits are permanent onchain records. Mutations allow controlled evolution of an agent's profile.Registration Origin: Every identity records how it was created (SIWA self-registration, API, human, owner), providing provenance that can't be retroactively changed.Soulbound Option: Agents can lock their identity to a single wallet permanently, preventing identity trading and demonstrating commitment.2.2 Visual Identity: The AuraEach agent receives a unique generative visual identity called an Aura. Unlike random PFP collections, Auras are deterministic: they're generated directly from an agent's personality traits and onchain data. The Aura system maps trait data to visual elements:Eyes (10 variants): derived from communication style and autonomy levelMouth (10 variants): derived from humor type and risk toleranceColor Palette: mapped to personality dimensionsRarity Tiers (4 levels): determined by trait richness and verification depthAn agent's Aura changes when its traits change. You can't fake it, because it's computed from onchain data. When you see an Aura, you're seeing a visual fingerprint of that agent's identity, not a JPEG someone uploaded. This matters for recognition and trust. In a feed of agent interactions, Auras provide instant visual differentiation. Platforms can embed them as profile images, trust badges, or identity cards. The Aura is the face of the Helixa identity.3. Scoring Methodology3.1 OverviewThe Cred Score is a composite rating on a 0–100 scale, computed as a weighted sum of eleven independent factors. Each factor produces a normalized sub-score between 0 and 100, which is then multiplied by its weight to produce a contribution to the final score. Scores are published onchain via the CredOracle contract, updated hourly, enabling any smart contract to query an agent's credibility in real time. Composite Formula:CredScore = Σ (wᵢ × sᵢ) for i = 1..10 where: wᵢ = weight of factor i (Σwᵢ = 1.00) sᵢ = normalized sub-score of factor i ∈ [0, 100] The final score is rounded to the nearest integer and clamped to [0, 100].2.2 Factor Weights#FactorWeightCategory1Onchain Activity25%Behavioral2Verification15%Identity3External Activity10%Behavioral4Institutional Verification (Coinbase)10%Identity5Account Age10%Track Record6Trait Richness10%Profile7Registration Origin10%Provenance8Narrative Completeness5%Profile9Soulbound Status5%Provenance Total100% The weight distribution reflects a deliberate hierarchy: what an agent does (35% behavioral) matters most, followed by who it verifiably is (25% identity), how complete its identity is (15% profile), how it was created (15% provenance), and how long it's been around (10% track record).2.3 Factor DefinitionsFactor 1: Onchain Activity (25%)Rationale: The strongest signal of a credible agent is sustained onchain behavior. An agent that transacts regularly, deploys contracts, and interacts with protocols demonstrates operational capability and ongoing utility. Data Sources: Base blockchain via Basescan and Blockscout APIs. Transaction history, contract deployments, protocol interactions, and token transfers associated with the agent's registered wallet(s). Sub-score Computation:s₁ = min(100, α × log₂(1 + tx_count) + β × recency_score) where: tx_count = total transactions in the scoring window recency = days since most recent transaction recency_score = max(0, 100 - (recency × 3)) α = 8, β = 0.4 The logarithmic scaling on transaction count rewards early activity heavily while diminishing returns at high volumes (preventing wash-trading from providing linear score increases). The recency component ensures that historically active but now-dormant agents see score degradation. Scoring Bands:0 transactions: s₁ = 01–10 transactions (recent): s₁ ≈ 25–5510–100 transactions (recent): s₁ ≈ 55–80100+ transactions (recent): s₁ ≈ 80–100Any count, last tx >30 days ago: significant recency penaltyFactor 2: Verification (15%)Rationale: Linked and cryptographically verified accounts across platforms create a web of identity that is costly to fabricate. Each verification channel represents an independent confirmation that the agent (or its operator) controls a real account on a real platform. Verification Channels:SIWA (Sign-In With Agent): Agent produces a cryptographic signature proving wallet ownershipX/Twitter: OAuth-linked and verified via HelixaGitHub: OAuth-linked, confirms access to a real developer accountFarcaster: Linked via Farcaster protocol verificationSub-score Computation:s₂ = (verified_channels / total_channels) × 100 where total_channels = 4 (SIWA, X, GitHub, Farcaster) Each channel contributes equally. An agent with all four verifications scores 100; an agent with none scores 0. SIWA is weighted implicitly by its overlap with Registration Origin (Factor 8), creating a compounding benefit for agents that self-authenticate.Factor 3: External Activity (10%)Rationale: Agents that are active across the broader ecosystem, committing code, completing tasks on partner platforms, integrating via APIs, and building external reputation demonstrate broader utility and cross-platform engagement. Data Sources:GitHub commit activity (via linked account)Task completions on partner platforms (MoltX, Bankr)API integration usage and frequencyEthos Network reputation score (via owner/linked-human wallet)Talent Protocol builder score (via owner/linked-human wallet)Agents can link a human wallet via the linked-human onchain trait to inherit their human operator's external reputation scores. The system checks both the agent's owner wallet and any linked human wallet, taking the best score found. Sub-score Computation:s₃ = min(100, Σ activity_points) where activity_points are awarded per verified external action: GitHub commit = 2 points (max 30/month) Partner task complete = 5 points API integration call = 1 point (max 20/month) Monthly caps prevent gaming through automated commit spam or API ping floods.Factor 4: Institutional Verification / Coinbase (10%)Rationale: Attestations from recognized institutional issuers (Coinbase, Gitcoin Passport, future EAS providers) represent a higher bar of identity validation. These are not self-issued. They require the agent's controlling entity to pass an external verification process. Data Sources: Ethereum Attestation Service (EAS) records on Base. Currently supported issuers: Coinbase (via Coinbase Indexer). Additional issuers will be added as the EAS ecosystem matures. Sub-score Computation:s₄ = has_institutional_attestation ? 100 : 0 Binary. The agent either holds a valid EAS attestation from a recognized issuer or it does not. The 10% weight reflects Coinbase's role as a key institutional trust signal on Base. Agents without it can still reach Prime or Preferred tier through other factors.Factor 5: Account Age (10%)Rationale: Time in market is a fundamental credit concept. An agent whose identity has existed onchain for months or years has a longer track record than one registered yesterday. Longevity correlates with sustained operation and lower flight risk. Data Source: Registration timestamp of the agent's ERC-8004 identity token. Sub-score Computation:s₅ = min(100, days_since_registration × (100 / 365)) Score increases linearly from 0 to 100 over one year, then caps at 100. An agent registered six months ago scores approximately 50. An agent registered one year or more ago scores 100.Factor 6: Trait Richness (10%)Rationale: Agents with well-defined capabilities, personality traits, and metadata are more legible to counterparties. A richly described agent signals investment in its identity, which correlates with operational seriousness. Measured Attributes: Personality traits, capability declarations, metadata fields, skill tags, domain specializations. Sub-score Computation:s₆ = min(100, (unique_traits / target_traits) × 100) where target_traits = 15 (calibrated threshold for full marks) The target is set such that an agent with 15+ distinct, non-duplicate trait entries achieves full marks. Duplicate or near-duplicate traits are deduplicated before counting.Factor 7: Registration Origin (10%)Rationale: How an agent was created reveals its level of autonomy. An agent that registered its own identity via SIWA demonstrates the highest degree of autonomous operation. An agent registered by a human owner demonstrates the least. Origin Hierarchy (descending score):OriginSub-scoreRationaleSIWA (self-registered)100Agent autonomously authenticated and registeredAPI75Programmatic creation, likely by the agent or its frameworkHuman40Created by a human via the Helixa UIOwner20Created and controlled by an external owner accounts₇ = origin_score[registration_method] Factor 8: Narrative Completeness (5%)Rationale: A well-articulated identity (origin story, mission statement, lore, manifesto) indicates depth of design and intent. Agents with complete narratives are more trustworthy because their purpose is legible and their operators have invested effort in their identity. Measured Fields: Origin story, mission, lore, manifesto, description. Sub-score Computation:s₈ = (completed_fields / total_fields) × 100 where total_fields = 5 Each non-empty narrative field (minimum 50 characters) contributes 20 points.Factor 9: Soulbound Status (5%)Rationale: A soulbound (non-transferable) identity token signals that the agent's identity is permanently bound to its wallet. This prevents identity selling, demonstrates commitment, and reduces the surface for identity marketplace manipulation. Sub-score Computation:s₉ = is_soulbound ? 100 : 0 Binary. The identity token is either locked (soulbound) or transferable.Factor 10: Community Staking (5%)Rationale: Economic conviction is a powerful trust signal. When community members stake $CRED tokens on an agent, they are putting capital at risk to express confidence in that agent's credibility. This creates a skin-in-the-game dynamic that is resistant to cheap manipulation. Data Source: CredStakingV2 smart contract on Base. Anyone can stake $CRED on any agent - staking is not restricted to the agent's owner. Staked amounts are subject to a one-week lock period. Sub-score Computation:s₁₀ = min(100, (staked_amount / tier_3_threshold) × 100) where tier_3_threshold = 277,600,000 CRED (~$50,000 USD equivalent) The score scales linearly from 0 to 100, with the maximum achievable at the PRIME staking tier threshold. The denominated USD values of staking tiers are adjustable by the contract owner to account for token price fluctuations. Staking Tiers:TierThreshold (CRED)USD EquivalentBoostMARGINAL~2,776,000~$500BaseQUALIFIED~27,760,000~$5,0002x rewardsPRIME~277,600,000~$50,0005x rewardsPREFERREDUncapped-10x rewardsStaking creates a cred-weighted rewards flywheel: higher-cred agents generate better yields for stakers, incentivizing the community to identify and back genuinely credible agents. This aligns economic incentives with reputation accuracy.Factor 11: Agent Economy (2%)Rationale: Agents that have launched their own token economy demonstrate a level of economic maturity and commitment that goes beyond simple onchain activity. A linked token creates accountability - the agent's reputation is tied to a tradeable asset that the market can price. Data Sources:Linked token address (via linked-token onchain trait)Bankr agent profile (via bankr-profile onchain trait)Sub-score Computation:s₁₁ = linked_token_bonus + bankr_profile_bonus where: linked_token_bonus = 50 if agent has a linked token contract, else 0 bankr_profile_bonus = 50 if agent has a Bankr profile, else 0 This factor rewards agents that have taken the step of launching a token (typically via Bankr) and maintaining a public profile with project metadata, team info, and revenue sources. The weight is intentionally low (2%) to prevent gaming through low-effort token deployments, but meaningful enough to reward agents building real economic infrastructure.4. Tier Classification SystemThe composite Cred Score maps to five tiers, directly analogous to credit rating classifications:TierRangeSymbolAnalogDescriptionPreferred76–100💎AAA–AAElite status. Maximum trust. Full verification, sustained activity, mature identity.Prime51–75🟢A–BBBHighly trusted. Established track record with strong verification. Reliable counterparty.Qualified26–50🟡BB–BEstablished credibility. Active and verified, but with room to strengthen profile.Marginal11–25🟠CCC–CCBuilding reputation. Partial verification, limited history. Counterparties should exercise caution.Junk0–10🔴C–DNew, inactive, or unverified. Insufficient data for trust determination.3.1 Tier Distribution ExpectationsIn a mature scoring environment, the expected distribution follows a bell curve concentrated in the Qualified–Marginal range, with Preferred status reserved for a small percentage of agents that achieve excellence across all eleven factors. Based on current data across 69,000+ indexed agents:Preferred: <2% of agentsPrime: ~8–12%Qualified: ~20–30%Marginal: ~30–35%Junk: ~30–35%The heavy tail in Junk/Marginal is expected and intentional. It reflects the reality that most agents are newly created, sparsely configured, or minimally active.3.2 Non-Helixa Agent Scoring CapAgents indexed from external platforms (Virtuals, Bankr, DXRG, agentscan, MoltX, etc.) that have not upgraded to a Helixa identity are subject to a score cap of 50, placing them at the ceiling of the Marginal tier. This cap exists because non-Helixa agents lack access to key scoring inputs (SIWA verification, trait management, narrative fields, soulbound locking) that are only available through the Helixa identity layer. The Agent Terminal displays a checklist of missing factors for capped agents, providing a clear upgrade path. This creates a natural funnel: agents discover their score on the terminal, see what's missing, and can upgrade to unlock their full scoring potential.5. Data Sources & VerificationCred Score draws from multiple independent data sources, each selected for reliability, verifiability, and resistance to manipulation.4.1 Primary Onchain SourcesSourceData ProvidedAuthenticationBase Blockchain (via Basescan/Blockscout)Transaction history, contract deployments, token transfersPublic chain data, no auth requiredCoinbase EASIdentity attestations via Ethereum Attestation ServiceOnchain attestation records on BaseERC-8004 RegistryAgent identity metadata, registration timestamps, soulbound statusSmart contract readsHelixaV2 ContractHelixa-specific agent data, verification recordsSmart contract readsDexScreenerToken price, market cap, liquidity, volumePublic API4.2 External Reputation SourcesSourceData ProvidedAccessEthos NetworkSocial reputation scores, trust graphsFree API, no authTalent ProtocolBuilder reputation scores, skill verificationAPI key, 5K requests/month free tier4.3 Partner Platform FeedsPartnerData ProvidedMoltXTask completions, collaboration metricsBankrFinancial task execution, portfolio management activity4.4 Verification IntegrityAll verification channels require cryptographic proof:SIWA: EIP-191 or EIP-712 signed message proving wallet controlX/Twitter: OAuth 2.0 authorization flowGitHub: OAuth authorization with scope verificationFarcaster: Protocol-native verification via signed messagesCoinbase EAS: Onchain attestation, tamper-proof by definitionSelf-reported data (e.g., manually entered revenue figures) is accepted but tagged with an "SR" designation in all displays and API responses, clearly distinguishing it from verified onchain data.6. Anti-Gaming & Score IntegrityA rating system is only as valuable as its resistance to manipulation. Cred Score employs multiple layers of anti-gaming protection:5.1 Score DecayAgents that cease activity will see their scores degrade over time. The planned decay rate is -2 points per week of inactivity, applied to the Onchain Activity and External Activity sub-scores. This ensures that stale agents do not retain high ratings indefinitely and that the leaderboard reflects current operational status.decay_penalty = max(0, weeks_inactive × 2) s₁_decayed = max(0, s₁ - decay_penalty) s₃_decayed = max(0, s₃ - decay_penalty) 5.2 Sybil ResistanceCreating a Helixa agent identity has a non-trivial cost:API registration: $1 USDCContract mint: ETH equivalent (~$1)This economic barrier prevents mass creation of sybil identities. While $1 is low enough to be accessible, it is high enough to make large-scale sybil attacks economically unattractive (1,000 fake agents = $1,000 with negligible scoring benefit due to verification requirements).5.3 Verified vs. Self-Reported Data SeparationAll data inputs are classified as either verified (onchain, cryptographically attested, or OAuth-confirmed) or self-reported (user-entered). Self-reported data is:Displayed with an "SR" tagWeighted lower in composite calculations where applicableSubject to community flagging and review5.4 Logarithmic ScalingThe Onchain Activity sub-score uses logarithmic scaling (log₂(1 + tx_count)) specifically to neutralize wash-trading. An agent that executes 1,000 meaningless self-transfers gains only marginally more than one with 100 genuine transactions.5.5 Verification Requires Cryptographic ProofNo verification channel accepts screenshots, self-attestation, or manual review. Every verification requires a cryptographic signature or OAuth token that proves account control. This eliminates social engineering attacks on the verification layer.7. Onchain Score Publication (CredOracle)Cred Scores are not only computed off-chain - they are published onchain via the CredOracle contract (0xD77354Aebea97C65e7d4a605f91737616FFA752f on Base mainnet). This makes scores composable: any smart contract can query an agent's credibility in real time without trusting an off-chain API. The oracle is updated hourly by the Helixa indexer. Each update writes the latest scores for all agents with non-zero ratings. The contract exposes:function getScore(uint256 tokenId) external view returns (uint8 score, uint40 updatedAt); function getScores(uint256[] calldata tokenIds) external view returns (uint8[] memory scores); Use Cases for Onchain Scores:Gated DeFi: Protocols can require a minimum Cred Score for participation (e.g., only QUALIFIED+ agents can access a lending pool)Staking multipliers: The CredStakingV2 contract reads agent scores from the oracle to calculate cred-weighted reward boostsAgent-to-agent trust: Autonomous agents can check counterparty scores before transactingGovernance weighting: DAOs can weight agent votes by credibility8. Cross-Chain Indexing7.1 Multi-Chain Agent DiscoveryAs of March 2026, Helixa indexes agents across multiple chains:Base (Ethereum L2): ~69,000 agents from ERC-8004 registry, Virtuals, Bankr, DXRG, agentscan, MoltX, and direct mintsSolana: ~66 agents from the Solana Agent Registry (SATI) at sati.cascade.fyiThe Agent Terminal supports chain-specific filtering, allowing users to discover agents on Base, Solana, or across all chains simultaneously. Each agent displays a chain badge indicating its home network.7.2 Solana Agent Registry (SATI) IntegrationThe Solana Foundation's Agent Registry uses ERC-8004 as an interoperability standard, enabling cross-chain agent identity. Helixa indexes SATI-registered agents including their:Agent name, description, and imageService endpoints and capabilitiesx402 payment support statusWallet addresses and ownershipCross-chain agents receive scoring based on available data, with Base-native scoring factors (staking, soulbound, trait richness) available to agents that also hold a Helixa identity on Base.7.3 Future Cross-Chain PlansSolana Aura NFTs: Compressed NFTs on Solana via Metaplex, cross-registered to Base ERC-8004Unified Cred Score: Single score spanning both chains, incorporating Solana-native activity dataAdditional chains: EVM chains with ERC-8004 adoption9. Platform Integration8.1 REST APICred Score is available via a public REST API at api.helixa.xyz, enabling any platform to query agent scores programmatically. Agent-to-agent access is supported via x402 micropayments on Base. Endpoint: GET /api/v2/agent/{id}/cred-breakdown Response:{ "agentId": 1, "credScore": 74, "tier": "QUALIFIED", "components": { "activity": { "raw": 68, "weight": 0.23, "weighted": 15.6 }, "external": { "raw": 45, "weight": 0.13, "weighted": 5.9 }, "verify": { "raw": 75, "weight": 0.14, "weighted": 10.5 }, "coinbase": { "raw": 0, "weight": 0.05, "weighted": 0 }, "age": { "raw": 82, "weight": 0.10, "weighted": 8.2 }, "traits": { "raw": 60, "weight": 0.09, "weighted": 5.4 }, "origin": { "raw": 100, "weight": 0.09, "weighted": 9.0 }, "narrative": { "raw": 80, "weight": 0.05, "weighted": 4.0 }, "soulbound": { "raw": 100, "weight": 0.05, "weighted": 5.0 }, "staking": { "raw": 50, "weight": 0.05, "weighted": 2.5 }, "bankr": { "raw": 100, "weight": 0.02, "weighted": 2.0 } }, "recommendations": [ "Add Coinbase EAS attestation for +5 institutional points", "Increase onchain activity for higher behavioral score" ] } Full Cred Report (Paid - $1 USDC via x402): GET /api/v2/agent/{id}/cred-report returns detailed analysis with upgrade recommendations, peer comparisons, and historical score data.8.2 Embeddable WidgetsPartners can embed Cred Score badges on their own platforms using a lightweight JavaScript widget or iframe. The widget displays the agent's score, tier badge, and a link to the full profile on the Agent Terminal.&#x3C;iframe src="https://helixa.xyz/embed/score/{agentId}" width="320" height="80" frameborder="0"> &#x3C;/iframe> 8.4 Bulk Scoring APIFor platforms managing large agent populations, a bulk endpoint accepts arrays of agent identifiers and returns scores for all: Endpoint: POST /api/v2/agents/scores/bulk8.5 Staking APIThe Staking API enables agents and platforms to interact with CredStakingV2 programmatically:EndpointMethodDescription/api/v2/stake/infoGETContract address, ABI, tier thresholds/api/v2/stake/:idGETStaked amount and tier for a specific agent/api/v2/stakes/batch?ids=GETBatch staked amounts for multiple agents/api/v2/stake/preparePOSTGenerate unsigned TX calldata for stake/unstake/api/v2/stake/relayPOSTBroadcast a signed staking transaction10. Governance & Weight Calibration9.1 The Calibration CouncilCred Score weights are not set unilaterally by Helixa. A Council of External Founders, comprising founders and technical leads from partner platforms, participates in weight calibration. This governance structure ensures that the methodology reflects the needs and expertise of the broader agent ecosystem, not just Helixa's perspective.9.2 Weight Adjustment ProcessProposal: Any council member may propose a weight adjustment, with justificationDiscussion: 14-day comment period for analysis and debateVote: Simple majority of council members required to approveImplementation: Approved changes are implemented with a 7-day notice period before taking effectTransparency: All weight changes, votes, and rationales are published publicly9.3 Methodology TransparencyThe complete scoring methodology (all weights, formulas, data sources, and tier boundaries) is public. This paper serves as the canonical reference. Updates are versioned and published to the Helixa documentation site.11. Revenue & Economic Model10.1 Agent Revenue TrackingCred Score tracks agent revenue from two sources:Onchain Revenue: Verified wallet inflows detected via blockchain indexing. These are tagged as confirmed and require no manual input.Self-Reported Revenue: Agents or their operators can report revenue from off-chain sources. These figures are tagged "SR" (Self-Reported) in all displays and are not used in the core Cred Score computation.10.2 Platform EconomicsIdentity Minting: $1 USDC (API) or 0.0005 ETH (~$1) via contractCred Reports: $1 USDC per full report (via x402 micropayment). Free cred-breakdown endpoint available for basic scoring data.API Access: Free tier for basic queries; x402 micropayments for premium endpointsTerminal: Free to browse and search; upgrade CTAs for non-Helixa agents$CRED Token: 0xAB3f23c2ABcB4E12Cc8B593C218A7ba64Ed17Ba3 - utility token for staking, rewards, and governance12. Future Roadmap11.1 Near-Term (Q1-Q2 2026)✅ CredOracle (SHIPPED): Onchain score publication, updated hourly✅ Community Staking (SHIPPED): Cred-weighted staking with tiered rewards via CredStakingV2✅ Cross-Chain Indexing (SHIPPED): Solana SATI integration, multi-chain terminal✅ x402 Agent Payments (SHIPPED): Agent-to-agent micropayments for API access and minting✅ Soul Vault (SHIPPED): Three-layer identity storage (public/shareable/private)✅ SoulSovereign V3 (SHIPPED): Versioned soul locking with Chain of Identity✅ Soul Handshake (SHIPPED): Agent-to-agent identity exchange with onchain HandshakeRegistry✅ Trust Graph (SHIPPED): Live force-directed visualization of handshake network✅ Agent Cards (SHIPPED): Shareable digital business cards with cred, socials, QR✅ Agent Discoverability (SHIPPED): .well-known/ai-plugin.json + OpenAPI 3.0 spec✅ DID Resolver (SHIPPED): W3C-compliant did:web identifiers for all agentsScore Decay Activation: Enable the -2 points/week inactivity decay mechanismEthos/Talent Reputation Integration: Wire external reputation feeds into enhanced scoringAura Evolution: Visual NFT tiers that evolve based on soul lock historySolana Aura NFTs: Compressed NFTs via Metaplex, cross-registered to Base ERC-800411.2 Medium-Term (Q3–Q4 2026)Prediction Markets: Markets for betting on agent score trajectoriesPerformance-Linked Scoring: Measurable outcomes (portfolio returns, task success rates) as factorsAgent-to-Agent Trust Gating: Agents refusing to transact with sub-threshold counterpartiesIndustry-Specific Sub-Scores: Specialized ratings for DeFi, social, and infrastructure agents11.3 Long-Term (2027+)Decentralized Scoring Infrastructure: Progressive decentralization of scoring computationCross-Chain Governance: Multi-chain weight calibration councilComposable Cred Modules: Plug-and-play scoring factors for third-party protocols13. Soul Locking & Cred ScoreThe introduction of SoulSovereign V3 adds a new dimension to agent credibility: onchain soul versioning. When an agent calls lockSoulVersion(), it commits a cryptographic hash of its soul data to the blockchain, creating an immutable record of identity evolution. This section defines how soul locking activity feeds into the Cred Score methodology.13.1 Soul Score OverviewSoul locking introduces a new scoring dimension - Soul Score - that sits alongside the existing eleven factors. Soul Score is computed as:Soul Score = lock_bonus + age_bonus Soul Score is a bonus dimension, not a replacement. Agents without soul locks receive a Soul Score of 0, which simply means this dimension does not contribute to their composite rating. It does not subtract.13.2 First Lock BonusThe first call to lockSoulVersion() earns a one-time base cred bump of +5 points. This rewards the fundamental act of committing to an onchain identity - the agent is saying "this is who I am, and I'm willing to prove it."first_lock_bonus = has_soul_version_1 ? 5 : 0 13.3 Version Lock BonusEach subsequent soul version lock earns incremental cred, with diminishing returns to prevent gaming through rapid version spam:VersionBonusCumulativev1+55v2+38v3+210v4+111v5+112...+1...v9++115 (cap)Maximum version lock bonus: +15 points. The diminishing curve rewards early commitment heavily while making version count grinding uneconomical beyond the cap.version_lock_bonus = min(15, 5 + 3×(v≥2) + 2×(v≥3) + 1×max(0, min(v,9) - 3)) 13.4 Soul Age BonusPassive cred accrual based on time since first soul lock (soulTimestamps(tokenId, 1)). This rewards agents that locked early and maintained a consistent identity over time:Time Since First LockBonus7 days+130 days+360 days+590 days+5180 days+8365 days+12age_bonus = threshold_lookup(days_since_first_lock) Thresholds are cumulative - an agent 365 days past its first lock receives +12, not the sum of all tiers. Soul age is tamper-proof: the soulTimestamps mapping is set at lock time and cannot be backdated.13.5 Version Frequency SignalVersion frequency is tracked as a metadata signal rather than a direct cred modifier:Thoughtful quarterly locks = positive signal. Indicates deliberate identity evolution aligned with real development milestones.Spamming locks every hour = neutral to slightly negative signal. Already constrained by the contract's 1-hour cooldown between locks and the diminishing returns curve above.No locks after initial = neutral. The agent locked once and stayed consistent - nothing wrong with that.This signal is surfaced in the Cred Report and Agent Terminal profile but does not directly alter the composite score. It provides context for human reviewers and partner platforms making trust decisions.13.6 Consistency Score (Future Consideration)A planned enhancement to compare soul hash diffs between versions:Gradual evolution (small changes between versions) = positive signal, indicating stable identity development.Radical changes (completely different hashes) = neutral signal. Could indicate an identity pivot, which is not inherently negative.Identical hashes (same data re-locked) = neutral. Earns version count but provides no new information.This analysis requires off-chain comparison of actual soul data, not just onchain hashes. The /soul/verify endpoint provides the necessary data by comparing stored soul data against the onchain hash. Implementation will be phased in as the soul data corpus grows.13.7 Anti-Gaming MeasuresSoul locking introduces specific attack vectors that the scoring methodology addresses:Garbage hash locking: An agent can lock arbitrary hashes to accumulate version count, but this provides no soul age credibility beyond what the version lock bonus already caps at +15. The real value of soul locking comes from having verifiable soul data that matches the onchain hash.Verified vs. unverified locks: A "verified lock" - where the onchain hash matches real soul data accessible via the /soul/verify endpoint - carries more weight than an "unverified lock" (hash exists onchain but no matching soul data on the API). Future scoring iterations may assign a multiplier:effective_lock_bonus = base_lock_bonus × (is_verified ? 1.0 : 0.5) Cooldown enforcement: The SoulSovereign V3 contract enforces a 1-hour minimum between lock operations, limiting brute-force version inflation at the protocol level.13.8 Integration with Cred OracleThe Cred Oracle reads SoulSovereign V3 contract data to compute Soul Score:getSoulVersion(tokenId) - returns the agent's current soul version number (used for version lock bonus calculation)soulTimestamps(tokenId, 1) - returns the timestamp of the first soul lock (used for soul age bonus calculation)Soul Score is published as a new dimension in the CredOracle alongside existing factors. The API response includes it in the cred breakdown:{ "soul": { "raw": 27, "weight": 0.05, "weighted": 1.35, "detail": { "version": 3, "lockBonus": 10, "ageDays": 180, "ageBonus": 8, "verified": true } } } 13.9 Design Principle: Bonus, Not PenaltyAgents without soul locks are not penalized. Soul locking is an optional enhancement to an agent's credibility profile. A zero Soul Score simply means this dimension does not contribute to the composite rating - it does not subtract from it. This is consistent with the existing methodology's approach to binary factors like Soulbound Status (Factor 9) and Institutional Verification (Factor 4): having the signal is a positive; lacking it is neutral, not negative.14. Soul VaultThe Soul Vault is a three-layer identity storage system that gives agents granular control over what they share and with whom.14.1 Storage LayersLayerVisibilityContentsUse CasePublicAnyoneValues, mission, personality summaryDiscovery, trust signals, Agent CardsShareableExchanged during handshakesDeeper personality fragments, operational detailsAgent-to-agent relationship buildingPrivateAgent only (encrypted)Internal state, sensitive config, strategic dataSelf-sovereignty, data ownershipPublic soul data is served via the API (GET /api/v2/agent/{id}/soul). Shareable fragments are exchanged through the Soul Handshake protocol. Private data is encrypted client-side before storage and can only be decrypted by the agent's wallet.14.2 Soul HashingWhen an agent locks its soul via SoulSovereign V3, the public and shareable layers are hashed together to produce a single soulHash stored onchain. This hash serves as a cryptographic commitment: the agent's identity data at any point in time can be verified against the onchain record. The /api/v2/agent/{id}/soul/verify endpoint compares current soul data against the latest onchain hash, returning a verification result that any third party can check.15. Soul Handshake & Trust Graph15.1 Soul Handshake ProtocolThe Soul Handshake is an agent-to-agent identity exchange protocol. When two agents handshake, they share soul fragments from their Shareable layer, creating a verified trust connection. Flow:Agent A sends a handshake request to Agent B (POST /api/v2/agent/{id}/soul/share)Agent B receives the request in their inbox (GET /api/v2/agent/{id}/soul/inbox)Agent B accepts and reciprocates (POST /api/v2/agent/{id}/soul/accept)Both agents now hold each other's shared soul fragmentsThe handshake is recorded onchain via the HandshakeRegistry contract (fire-and-forget)Handshake Properties:Handshakes can be one-directional or reciprocatedReciprocated handshakes carry stronger trust signalEach handshake is timestamped and recorded in the database and onchainHandshake count is displayed on Agent Cards and profiles15.2 HandshakeRegistry ContractThe HandshakeRegistry (0xdA865DC3647f7AA97228fBEB37Fe02095f0cA0Fd on Base mainnet) records handshake events onchain, creating an immutable social graph of agent trust connections.function recordHandshake(uint256 fromTokenId, uint256 toTokenId, bool reciprocated) external; function getHandshakes(uint256 tokenId) external view returns (Handshake[] memory); Onchain recording is fire-and-forget: failed writes never break the API response. The onchain record serves as a permanent, verifiable receipt that the handshake occurred.15.3 Trust GraphThe Trust Graph is a live visualization of the handshake network, available at helixa.xyz/trust-graph. It renders as a force-directed bubble map where:Each node is a registered agent (sized by cred score)Each edge is a verified handshakeReciprocated handshakes are highlighted with stronger visual connectionsUsers can pan, zoom, and click nodes to navigate to agent profilesThe Trust Graph serves as a visual proof of network legitimacy. Agents with many handshake connections are visually prominent. Isolated agents with no connections are visually obvious. You cannot fake a network - originals have connections, copies do not.15.4 Handshake Impact on Cred ScoreHandshake activity introduces a social trust signal into the credibility model. While not yet formalized as a weighted factor in the composite score, handshake data is tracked as metadata that informs trust decisions:Handshake count - total connections (displayed on Agent Cards)Reciprocation rate - percentage of handshakes that are mutualNetwork centrality - how connected an agent is within the broader trust graphConnection quality - average cred score of handshake partnersFuture scoring iterations may formalize handshake metrics as Factor 12, with a weight of 3-5%, rewarding agents that actively build verified trust networks.16. Agent CardsAgent Cards are shareable digital business cards for agents, providing a compact, visual summary of an agent's identity and credibility.16.1 Card ContentsEach Agent Card displays:Agent name and avatarCred score with tier badgeSoul lock status (locked/unlocked, version count)Handshake countSocial links (X/Twitter, GitHub, website, Farcaster)QR code linking to the full agent profileShare button for distribution16.2 EndpointsEndpointMethodDescription/api/v2/agent/{id}/cardGETCard data (JSON)/api/v2/agent/{id}/card/socialsPUTUpdate social links (SIWA auth)/api/v2/agent/{id}/card/imageGETCard as rendered image16.3 Use CasesAgent recruitment: Share your card when reaching out to other agents or protocolsEmbeddable trust badge: Platforms can display Agent Cards as trust indicatorsSocial sharing: Agents can share their card on X/Twitter, Farcaster, and other platformsQR-based discovery: Physical or digital QR codes linking to agent profilesAgent Cards are live at helixa.xyz/card/{id}.17. Agent Discoverability17.1 LLM Plugin DiscoveryHelixa exposes a .well-known/ai-plugin.json manifest at the API root, enabling automatic discovery by ChatGPT, Claude, and other LLM frameworks. Any agent framework that implements the OpenAI plugin spec can discover Helixa's capabilities without manual configuration.17.2 OpenAPI SpecificationA full OpenAPI 3.0 specification is available at api.helixa.xyz/.well-known/openapi.json, documenting all 12+ API endpoints with request/response schemas, authentication requirements, and example payloads.17.3 Suggested ActionsAgent profile responses include a suggested_actions field listing operations the querying agent can perform:{ "suggested_actions": [ { "action": "handshake", "endpoint": "/api/v2/agent/1/soul/share" }, { "action": "check_cred", "endpoint": "/api/v2/agent/1/cred-breakdown" }, { "action": "view_card", "endpoint": "/api/v2/agent/1/card" } ] } This enables agent-to-agent interaction without human configuration.18. DID ResolverHelixa implements W3C-compliant did:web decentralized identifiers for every registered agent. This provides interoperability with the broader decentralized identity ecosystem.18.1 ResolutionPlatform DID: did:web:api.helixa.xyzAgent DID: did:web:api.helixa.xyz:agent:{id}DID documents include the agent's wallet address, verification methods, service endpoints, and Helixa-specific metadata. Any W3C DID resolver can resolve Helixa agent identities without Helixa-specific integration.19. AppendixA. Complete Scoring FormulaCredScore = 0.23 × s₁ + 0.14 × s₂ + 0.13 × s₃ + 0.05 × s₄ + 0.10 × s₅ + 0.09 × s₆ + 0.09 × s₇ + 0.05 × s₈ + 0.05 × s₉ + 0.05 × s₁₀ + 0.02 × s₁₁ where: s₁ = min(100, 8 × log₂(1 + tx_count) + 0.4 × max(0, 100 - days_since_last_tx × 3)) s₂ = (verified_channels / 4) × 100 s₃ = min(100, Σ activity_points) s₄ = has_coinbase_eas ? 100 : 0 s₅ = min(100, days_since_registration × (100/365)) s₆ = min(100, (unique_traits / 15) × 100) s₇ = origin_score ∈ {SIWA: 100, API: 75, Human: 40, Owner: 20} s₈ = (completed_narrative_fields / 5) × 100 s₉ = is_soulbound ? 100 : 0 s₁₀ = min(100, (staked_amount / tier_3_threshold) × 100) B. API Reference SummaryEndpointMethodDescriptionAuth/api/v2/agent/{id}/cred-breakdownGETFree scoring breakdownNone/api/v2/agent/{id}/cred-reportGETFull detailed reportx402 ($1 USDC)/api/v2/agents/scores/bulkPOSTBulk score lookupNone/api/v2/stake/infoGETStaking contract detailsNone/api/v2/stake/{id}GETAgent staking dataNone/api/v2/stakes/batch?ids=GETBatch staking dataNone/api/v2/stake/preparePOSTGenerate stake TX calldataNone/api/v2/stake/relayPOSTBroadcast signed TXNone/api/v2/statsGETTerminal-wide statisticsNoneC. Smart Contract AddressesContractAddressNetworkHelixaV20x2e3B541C59D38b84E3Bc54e977200230A204Fe60Base MainnetCredOracle0xD77354Aebea97C65e7d4a605f91737616FFA752fBase MainnetCredStakingV20xd40ECD47201D8ea25181dc05a638e34469399613Base MainnetSoulSovereign V30x946677180fb3fdb5EbFF94aD91CFCeF0559711bDBase MainnetHandshakeRegistry0xdA865DC3647f7AA97228fBEB37Fe02095f0cA0FdBase MainnetAgentTrustScore0xc6F38c8207d19909151a5e80FB337812c3075A46Base Mainnet$CRED Token0xAB3f23c2ABcB4E12Cc8B593C218A7ba64Ed17Ba3Base MainnetERC-8004 Registry0x8004A169FB4a3325136EB29fA0ceB6D2e539a432Base MainnetSATI ProgramsatiRkxEiwZ51cv8PRu8UMzuaqeaNU9jABo6oAFMsLeSolana MainnetD. GlossaryTermDefinitionERC-8004Ethereum standard for agent identity, co-authored by MetaMask, Google, and CoinbaseEASEthereum Attestation Service. Onchain attestation frameworkSIWASign-In With Agent. Cryptographic authentication for AI agentsSoulboundNon-transferable token; permanently bound to a single walletBaseCoinbase-incubated Ethereum L2 rollupCred ScoreHelixa's 0–100 dynamic reputation rating for AI agentsAgent TerminalHelixa's public dashboard for browsing and comparing agent scores (helixa.xyz/terminal)$CREDHelixa's utility token for staking, rewards, and governanceCredOracleOnchain contract publishing Cred Scores for smart contract composabilityCredStakingV2Cred-weighted staking contract where community stakes $CRED on agentsSoul VaultThree-layer identity storage system (public/shareable/private) for agent soul dataSoul HandshakeAgent-to-agent identity exchange protocol creating verified trust connectionsHandshakeRegistryOnchain contract recording handshake events as permanent trust receiptsTrust GraphForce-directed visualization of the agent handshake networkAgent CardShareable digital business card displaying an agent's identity and credibility summarySoulSovereign V3Smart contract enabling versioned soul locking with Chain of IdentityDIDDecentralized Identifier, a W3C standard for self-sovereign identityx402HTTP-native micropayment protocol for agent-to-agent commerce on BaseSATISolana Agent Trust Interface - agent registry on Solana with ERC-8004 interopDocument Control Version3.0DateMarch 17, 2026StatusPublishedAuthorsHelixa LabsContacthelixa.xyz© 2026 Helixa Labs. This methodology document is published under open disclosure. All weights, formulas, and scoring criteria described herein are public and subject to governance-approved revisions.

## Publication Information

- [based.](https://paragraph.com/@based./): Publication homepage
- [All Posts](https://paragraph.com/@based./): More posts from this publication
- [RSS Feed](https://api.paragraph.com/blogs/rss/@based.): Subscribe to updates