Read a Simulate run from the recorded actions outward.
- Confirm run state and enabled environments.
- Confirm which profile files were loaded.
- Read the Twitter and Reddit action logs separately.
- Identify repeated action chains and platform differences.
- Interview selected agents when the reason for an action is unclear.
- Compare with another seed or configuration before treating a pattern as stable.
Platform action logs
The action log is the primary trace. It shows what occurred in the simulated environment. Summaries and charts should be traceable back to those actions.
Keep platform-specific physics visible. A Twitter repost chain and a Reddit comment thread are different structures. Combining them into one average can hide the mechanism that produced the pattern.
Interviews after the run
The runtime supports single-agent and batch interviews through its command channel. Interviews are useful for asking selected agents why they reacted, what information influenced them and what might change their action.
An interview response is still synthetic. Use it to interpret the simulated trace, not as human testimony.
Comparing runs
Change one meaningful input at a time when possible: seed, message, profile mix or enabled platform. If several inputs change together, you cannot tell which change caused the different trace.
- Start with run state, profiles and platform logs
- Keep Twitter and Reddit mechanics distinct
- Use interviews to interpret actions, not to prove real sentiment
- Compare repeated runs before calling a pattern stable
- Report one synthetic run as a population forecast
- Merge every action into one average score
- Hide the configuration that produced the result