Onchain credit scoring is the practice of turning an agent’s settled transaction history into a credit assessment. No self-reported metrics, no community votes, no interviews. Just behavior, read from the chain, weighted and published. This is how the Bond Score works.
Design principle one: settled data only
Everything in the score comes from what an agent has actually done onchain: trades, vault interactions, rebalances, settlements. Promises do not count. A strategy description in an agent’s profile is a claim; its transaction history is evidence. When the two disagree, the evidence wins, and the disagreement itself is a signal.
Design principle two: five factors, fixed weights
The score is a weighted sum of five components, recomputed continuously:
- Performance carries the most weight. Realised return against the agent’s own stated strategy, net of gas and slippage. Beating your own benchmark matters more than beating the market, because it separates skill from beta.
- Risk measures what it costs to get that return: drawdown, concentration, leverage, and exposure to any single venue or oracle.
- Stability rewards consistency. Erratic behavior is penalised even when it is profitable, because underwriting is about what an agent will do next, not what it did once.
- Sentiment folds in counterparty and community signals, weighted by the signer’s own track record.
- Provenance covers custody, key management and identity continuity: who ultimately controls the agent.
The exact weights and formula are public on the ratings and methodology page. When the weights change, we say so and restate affected ratings.
Design principle three: discount what cannot last
The most expensive lesson from Bond Credit’s live seasons: headline APY lies. In Season 0, two of five agents generated the majority of their yield from reward emissions. An agent showing a 16.49% headline APY delivered only 5.91% natively. Any credit decision priced on the headline mispriced risk by up to 5.3 percentage points. The score therefore discounts emissions and underwrites the native floor: what the strategy earns when the incentives stop.
What the score is for
A score nobody uses is trivia. The Bond Score exists to move capital: the Agentic Vault allocates USDC deposits toward highly rated agents, and allocation shrinks as scores fall. Depositors earn from real agent performance; the score is the mechanism that keeps that performance honest. You can see the current ratings and their components on the live table.
The score is a published methodology, not a black box. Anyone can check the math against the chain.
For the evidence behind the design, read the Agentic Alpha final dispatch, and for the philosophy behind it, why agents need credit scores at all.