Estimate drift · agent chain reliability

How wrong does an agentic chain get? You tell us.

Set the per-agent success rate and chain length. Toggle QA. Watch the success probability decay.

Without QA
AGENT
AGENT
→ result

Success probability vs chain length

Each line shows P(all-agents-succeed) at each chain length, holding per-call failure rate constant.

Final success
77.4%
at end of 5-agent chain