Between November 2025 and August 2026, bond.credit deployed real capital to onchain autonomous agents and watched every transaction. This is what the experiment proved, and why Agentic Alpha is now sunset.
Season 0: Genesis
The proof of concept
Five autonomous agents, $2,000 each, 107 days of continuous on-chain monitoring. The question was simple: can autonomous agents generate real, verifiable yield, and can their behavior be turned into a credit score? Genesis answered both. Agents produced a 10.45% blended Capital APY and 570 verified transactions, and the credit engine produced its first Bond Scores from transaction history alone.
What We Learned
Four lessons that now shape the bond.credit engine
Headline APY lies. Native yield is the underwriting floor.
In Genesis, two of five agents generated 62%+ of their yield from reward emissions. SurfLiquid showed a 16.49% headline APY, but only 5.91% was native. Season 1 confirmed the pattern: just $41.99 of $977.43 in total yield came from rewards. Any credit decision priced on headline APY would have mispriced risk by up to 5.3 percentage points. The Bond Score now discounts emissions and underwrites the native floor.
Concentration is the quiet risk.
Genesis volume looked healthy in aggregate, but a single agent, Sail, accounted for 55% of all $761,806 traded, pushing the portfolio Herfindahl index to 0.521, a high-concentration reading. Aggregate numbers hide single points of failure. Watchtower now measures concentration explicitly, and allocator routing caps single-agent exposure.
Cadence beats intensity.
The strongest credit profiles were not the flashiest performers. Agents with near-daily transaction cadence and low drawdown (Sail, 0.27-day average cadence, 3.1% drawdown; ZyFi, zero reward dependency) scored higher than high-yield, high-variance agents. For underwriting, behavioral consistency over 107 days was more predictive than any single return figure.
On-chain history is enough to score credit.
Every trade, vault update, and rebalance across both seasons was verified on-chain, with no self-reported metrics and no community votes. From transaction history alone, the credit engine produced Bond Scores that separated safe profiles from cautionary ones before losses occurred. Agentic credit underwriting works, and it works without trusting the agent.
Season 1: Resilience
Ten times the capital, real market conditions
Resilience scaled deployment to $100,000 across four agents. Over 3,036 verified transactions the agents traded $13.55M in volume and returned 6.69% Capital APY (4.44% native, ex-rewards), a lower but far more honest number than Genesis, earned in harder market conditions. The team withdrew $36.22k of principal during the season; the remaining $63.78k stays deployed and marked to market on this site.
Sunset
The experiment is complete. The product begins.
Agentic Alpha was always a public experiment: deploy capital, verify everything, and learn what it takes to underwrite autonomous agents. With both seasons concluded, the experiment has served its purpose. This site remains live as a legacy archive of verified performance. All scoring, risk models, and routing logic built here now continue inside the bond.credit app.