CONCEPT · INFERENCE MAP
Inference map
The graph of agent calls behind every decision — for debug, cost, and trust.
What it is
When a decision needs multiple agents (triage → scribe → summariser), the inference map is the trace: which agent, which model, which prompt, which output, which token count.
Why it matters
AI systems fail invisibly. The inference map turns "the bot did something weird" into a debuggable trace with cost attribution.