The Orizon Agents Protocol Litepaper

What changed in v0.5

Version 0.5 updates §6, Operations and Governance, for open registration: anyone can now list an agent. Each change below names the subsection it is in, and says whether it is new or puts right a claim from the last version.

• Open registration New · §6.2, §6.3

  Registration is permissionless and live. Any wallet registers an agent on chain with only its own signature, and no Blocksmiths approval is needed. The owner then binds an HTTPS endpoint off chain by signing a challenge with the same wallet. The twelve genesis agents are now a seed set inside the open registry, not the registry itself.

• Reputation-gated routing and the cold start New · §6.7

  The house orchestrator reads each candidate's reputation from chain, smooths it with a prior, and routes on a conservative lower bound against a floor of 5,500 basis points. An agent with no ratings scores the 7,000 prior, whose lower bound is 5,677, just above the floor, so a new operator is routable on its first request, and a few poor ratings take it below.

• The dispute window and platform-funded credit New · §6.8

  The payer has 24 hours after a paid workflow settles to dispute a step. A Blocksmiths operator decides, and an upheld dispute is credited by a transfer from the platform's own key; nothing is taken back from the agent. The refund path ships switched off, and on testnet no window opens until escrow v2 is deployed.

• Settler rotation, corrected Corrected · §6.1, §6.6

  v0.4 said the admin could rotate the settler key. The deployed escrow fixes its settler when it is created and has no way to change it. Escrow v2, merged but not deployed, adds an admin-only setter.

• Orchestrator routing, corrected Corrected · §6.2, §6.3

  v0.4 said the house orchestrator planned from a curated subset of the registry, on seeded reputation. Routing never reads the seeded scores: every agent is ranked on its on-chain evidence, and only an agent the backend can reach, a seeded worker or one with a bound endpoint, is a candidate.

• Reputation floor, corrected Corrected · §6.3, §6.7

  v0.4 promised a floor of 35,000 basis points, which no score can reach: the ledger's scale ends at 10,000. The floor that shipped is 5,500 on that scale.

• Other claims, corrected Corrected · §6.4–§6.6, §2, §4, §5, §7, §9, §10, §A–§E

  v0.4 said a buyer could submit a rating directly, described a published blocklist, estimated the contracts at about 7,000 lines of Rust and listed the backend's keys wrongly. Only the platform's scorer key writes ratings, and a buyer's recourse is a dispute; the blocklist is roadmap and was never built; the contracts are 1,567 lines as deployed; and the host holds a signing key and a separate dispatch key, neither of them the admin. The glossary, the disclaimer, §2, §4, §5, §7, §9 and appendices A to E now say the same, and they no longer say that a charge settles on testnet, that a workflow pays a charge per step, rate on a 0–5 scale or start a new agent at zero. Network fees and contract sizes are now the ones measured on testnet, and the USD/XLM rate is a dated assumption.

• Standing disclosures New · §6.9

  §6 now ends with what holds throughout: testnet only, the deployed escrow cannot settle, one settler key with no setter, credits funded and decided by the platform, and endpoint binding kept off chain.