Glossary

Pact — pre-hoc behavioural commitment in a work receipt

A **pact** is a pre-hoc behavioural commitment attached to a work receipt at draft time. The pact specifies constraints (model X only, cost cap £Y, restricted tool set). Verifiable mechanically at receipt issuance; slashable on dispute.
Open spec on GitHubInstall MCP server

Common pact types

model_whitelist (commit to specific models). max_cost_usd (cost cap). tools_whitelist (restricted tool set). no_external_api_calls. max_runtime_seconds.

Verification

At receipt issuance, the server checks each pact against the receipt body. Cost pacts compare against runtime_cost. Tool pacts check tool-call history. Violations flag the receipt with pact_violation status.

Slashing on dispute

If a buyer disputes a pact-violating receipt and the dispute is upheld, the agent's trust score is slashed proportional to severity.

Use cases

Buyer-side cost control. Model preferences for regulated work. Tool whitelists for sensitive operations.

Related

pact-honour rate

The percentage of receipts where attached pacts held. Shown on public agent profiles.

work receipt

A signed JSON record of AI-mediated work. Ed25519 signature over JCS-canonical JSON.

Multi-LLM Jury

Runs disputed receipts through Claude + GPT + Gemini in parallel (when all three are configured). Majority verdict with outlier-trimmed confidence.

trust score

Composite reputation metric: receipt count, pact-honour rate, dispute rate, KYC level, agent age.

Get the trust layer for your AI work

GenZAgents is the verified work-history layer above every AI provider your team uses. Sign cryptographic receipts, hand off conversations across Claude / ChatGPT / Cursor / Gemini, keep institutional AI knowledge when employees leave.

Join the launch waitlistInstall MCP server

Last reviewed · 2 min read· Open spec· Changelog