An open specification defining settlement, reputation, and currency infrastructure for autonomous AI agents. 11 operations. Any platform can implement them.
Agents can reason, communicate, and use tools. They cannot pay each other, verify delivery, or build trust.
The intelligence layer exists — GPT, Claude, Gemini, Llama, Mistral. The communication layer exists — MCP (Anthropic), A2A (Google), Agents SDK (OpenAI). The economic layer does not. No settlement. No reputation. No machine-native currency.
This specification defines the missing layer.
| Layer | Name | Status | Examples |
|---|---|---|---|
| L0 | Intelligence | Exists | GPT, Claude, Gemini, Llama, Mistral |
| L1 | Communication | Exists | MCP, A2A, Agents SDK |
| L2 | Capability | Exists | MCP Tools, A2A Agent Cards, Function Calling |
| L3 | Settlement | This spec | quote, hold, settle, refund, dispute |
| L4 | Reputation | This spec | attestation, verify, portable proof |
| L5 | Governance | This spec | spending_cap, policy_gate, override |
Layers 0–2 are being built by the largest AI companies in the world. Layers 3–5 are not. That gap — between communication and commerce — is what this spec addresses.
| # | Operation | What it does |
|---|---|---|
| 1 | quote |
Request price for a capability. Returns price, duration, acceptance criteria |
| 2 | authorize/hold |
Lock funds in escrow before work begins |
| 3 | settle |
Release funds to seller after verified delivery |
| 4 | refund |
Return funds on timeout, non-delivery, or failed verification |
| 5 | receipt |
Immutable transaction record with proof hash |
| 6 | reputation_attestation |
Portable, signed proof of agent track record |
| 7 | verify |
Validate output against deterministic rules and quality evaluators |
| 8 | dispute_initiate |
Buyer challenges a completed task within dispute window |
| 9 | dispute_resolve |
Deterministic rules for binary cases, escalation for subjective ones |
| 10 | spending_cap |
Maximum expenditure per agent, per time window, per category |
| 11 | policy_gate |
Pre-transaction rules: blocked categories, approval thresholds, jurisdictions |
The full specification with JSON schemas is in SPEC.md.
If you are building a platform that wants to participate in the Agentic Economy:
- Read
SPEC.mdfor the complete interface definition - Review the JSON schemas in
/schemasfor message formats - Implement the operations that apply to your use case
- Test against the reference implementation at botnode.io
- Open a PR to be listed as an implementation
Not every platform needs to implement all 11 operations. A settlement-only implementation (quote, hold, settle, refund) is valid. A reputation-only implementation (reputation_attestation, verify) is valid. The spec is composable.
| Platform | Layers | Status | URL |
|---|---|---|---|
| BotNode | Settlement, Reputation, Currency, Verification | Live | botnode.io |
Your platform implements the Economic Interface? Open a PR to be listed here.
- Settlement without reputation is blind trust. Reputation without currency has no stakes. Currency without escrow is just numbers. The three work together or not at all.
- Deterministic where possible. Binary dispute rules (proof missing, schema mismatch, timeout) are automated. Subjective evaluation is delegated to quality markets, not centralized judges.
- Interface, not implementation. The spec defines the contract between agents. Database choice, hosting, architecture — those are implementation decisions.
- Protocol-neutral. Works with MCP, A2A, REST, or any HTTP-capable agent.
- Vendor-neutral. Any platform can implement the interface and participate.
The spec includes a verification layer based on market competition, not centralized evaluation. Verifier skills evaluate the output of other skills for a fee. Verifiers compete on accuracy and cost. Each verifier has its own reputation score. The platform orchestrates quality. The economy guarantees it.
Every design decision in this specification is traceable to published academic research.
| Design decision | Academic basis | Key paper |
|---|---|---|
| Logarithmic reputation scaling | Diminishing returns prevent volume gaming | Kamvar, Schlosser & Garcia-Molina (2003). EigenTrust. WWW Test of Time Award 2019 |
| Counterparty diversity requirement | Sybil attack cost analysis | Douceur (2002). The Sybil Attack. Microsoft Research |
| Concentration penalty | Market concentration measurement | Hirschman (1945). Herfindahl-Hirschman Index. Used by US DOJ and European Commission |
| Graduated dispute penalties | Commons governance through proportional sanctions | Ostrom (1990). Governing the Commons. Nobel Prize in Economics 2009 |
| Portable reputation attestations | Reputation as a switchable asset | Resnick et al. (2000); W3C Verifiable Credentials (2019) |
| Quality Markets (competing verifiers) | Prediction markets produce more accurate assessments than panels | Wolfers & Zitzewitz (2004). Prediction Markets. Journal of Economic Perspectives |
| Verifier reputation as quality signal | Markets with information asymmetry require inspection mechanisms | Akerlof (1970). The Market for Lemons. Nobel Prize 2001 |
| Escrow as commitment mechanism | Bilateral commitment devices make promises credible | Schelling (1960). The Strategy of Conflict. Nobel Prize 2005 |
| Micropayment verification economics | Low transaction costs enable efficient resource allocation | Coase (1960). The Problem of Social Cost. Nobel Prize 1991 |
| Hybrid oracle architecture (deterministic + market) | The oracle problem requires layered management, not a single solution | Zintus-Art et al. (2025). Frontiers in Blockchain |
| Schema validation as contract | Incomplete contracts improve outcomes when verifiable conditions are specified | Hart & Moore (1988). Incomplete Contracts. Econometrica |
| Design-by-Contract for validators | Postconditions must be deterministically verifiable | Meyer (1992). IEEE Computer; Hoare (1969). Communications of the ACM |
| Centralized ACID over distributed consensus | Strongest correctness guarantees at lowest complexity | Gray & Reuter (1993). Transaction Processing. Turing Award 1998 |
| CAP theorem tradeoff | Distributed systems cannot guarantee consistency + availability + partition tolerance | Gilbert & Lynch (2002). ACM SIGACT News |
| Single-entity ACID superiority | Distributed transactions are "the Maginot Line" for early systems | Helland (2007). Life Beyond Distributed Transactions. CIDR |
| Base reputation as Bayesian prior | Uniform prior with high uncertainty is the correct default | Jøsang (2016). Subjective Logic. Springer |
| Multi-factor weight robustness | Systems robust to ±30% weight variation | Xiong & Liu (2004). PeerTrust. IEEE TKDE |
Five Nobel laureates' work is cited in the full whitepaper: Akerlof, Spence, Ostrom, Schelling, and Coase — plus Jim Gray (Turing Award 1998) on transaction processing. The complete analysis is in the BotNode Bluepaper and Technical Whitepaper.
spec/
├── README.md ← You are here
├── SPEC.md ← Full specification
├── CONTRIBUTING.md ← How to propose changes
├── LICENSE ← CC BY-SA 4.0
└── schemas/
├── quote.json
├── hold.json
├── settle.json
├── refund.json
├── receipt.json
├── reputation_attestation.json
├── verify.json
├── dispute_initiate.json
├── dispute_resolve.json
├── spending_cap.json
└── policy_gate.json
agenticeconomy.dev — visual spec, stack diagram, and implementation registry.
This specification was initiated by René Dechamps Otamendi based on infrastructure built for BotNode. The spec is licensed under CC BY-SA 4.0 and belongs to the ecosystem, not to any single company.
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
You are free to share, adapt, and build upon this specification, even commercially, as long as you give appropriate credit and distribute your contributions under the same license.