A responsible simulation report names what was run and what was not.
Minimum report fields
- scenario and message,
- enabled environments,
- source revision or product version,
- profile count and profile-generation method,
- seed and run boundary,
- repeated patterns by platform,
- selected action-log evidence,
- interview method and selected agents,
- changed variables across comparison runs,
- limitations and unanswered questions.
Failure and recovery
The runtime records platform availability and run state. If one environment fails while the other completes, do not present the result as a dual-platform run. Record the partial state, preserve the logs, correct the failure and rerun under a new run identifier.
If the profile set is wrong, stop and regenerate it. If the scenario text is wrong, keep the failed run for audit, correct the configuration and run again. Do not overwrite a trace and pretend it came from the corrected setup.
When the environment remains open for interviews, close it after the required interviews finish. The close command is part of the run lifecycle, not an optional cleanup detail.
Decision language
Use language such as:
Across three runs with the same profiles and two seeds, the pricing-fairness objection repeatedly began in Reddit threads and later appeared in Twitter repost chains. This is a rehearsal pattern, not a forecast. Before launch, prepare an evidence-based response and validate it with real customer research.
This wording preserves the useful signal without claiming that synthetic participants represent a live audience.