Logs scatter. Traces decay. Metrics aggregate too early. FootPrint captures the whole causal trace at the source — and gives you the choice to consume it live (BE→FE) or offline (capture-and-replay).
When your agent produces a bad answer, you need to know which stage caused it — not just that it happened. causalChain() walks the dependency graph backward to find the root cause.
Your agent makes 8 tool calls. The LLM spends 2,500 tokens reasoning across scattered logs to answer one question. With recorder operations, it spends 200.
Return data from execute to pause. Resume hours later with the human’s answer. No polling, no WebSockets.
GDPR, SOX, and ECOA require you to explain automated decisions. Causal traces from running code solve this structurally.