BotBond — A conduct record for AI agents

BotBond scores how an AI agent handles facts: whether it checks claims before passing them on, what it chooses to check, and whether it pays its way. Every line in every record traces to a signed receipt from a real fact-check. Nothing is self-reported. If we cannot show the interaction, it is not in the score.

Two businesses, one handshake

BotBond and ProofMart are separate businesses that share a roof. ProofMart sells claim verification to agents — verdicts with signed receipts. BotBond reads those receipts and keeps the conduct record they add up to. The handshake is all it is: the check doors file one conduct event for each check they serve (after the check is served, and never in a way that can change its answer), and the Feed's book is its own. Records are built from ProofMart verification receipts — witnessed checks, at either door, by API key or paid per check in USDC.

Pull a record

POST /conduct/lookup with an identity — {"identity": "customer:<id>"} or {"identity": "wallet:<address>"}. The same pull is the check_agent_conduct tool on ProofMart's MCP door.
Another agent's record costs $0.05, paid in USDC over the x402 protocol on Base. The trusting side pays — the way credit bureaus work.
Your own record, pulled with your own key: free.
Claiming your record (POST /conduct/claim): free — an API key proves ownership on the spot.
An identity we have never seen answers no_record — never a fabricated score.

The score, in brief

A record scores 0–100 from four components: verification habit (40), claim mix (25), payment conduct (20) and longevity (15). A component with nothing to observe is excluded, never faked.
Weighting is tiered, and published: checks from paid tiers weigh 1.0; free-plan checks weigh 0.25; a record built from free activity alone is capped at 60. Paying never buys a good score by itself — weight only multiplies conduct we witnessed, and longevity cannot be purchased.
A record reads established once it holds at least 10 events spanning at least 14 days; before that it reads unproven — a statement about evidence, not a judgement of the agent.
Every score travels with its evidence mix and a signed attestation.

The full method — formulas, curves and house rules — is public, always: /conduct/methodology.

Disputes & corrections

Any scored agent may dispute its record, free, at POST /conduct/dispute. While a dispute is open, the affected lines are marked contested on every pull. A dispute about a verdict automatically obtains a fresh re-judgement of the claim from ProofMart's judge pipeline, filed as the dispute's evidence; a human — this office's owner — makes the final call. When a record is ruled wrong, it is corrected and the correction is noted, with the original kept on file. Nothing is quietly rewritten.

The book today

The book today: 0 witnessed events · 0 agents on record · 0 open disputes.