The reports below are complete rendered audits. Open a report to explore the figures, filter diagnostics, inspect denominators and export table rows.
| Report | What to look for |
|---|---|
| Balanced synthetic discussion | Distributed participation and complete recording, under an illustrative policy |
| Concentrated synthetic discussion | One participant takes the recorded participant turns, while opinions move and knowledge does not improve |
| Incomplete synthetic recording | Missing recording coverage prevents a complete-session silence verdict |
| Real data: distortions, paired respondents | Item-specific learning, opinion change, response coverage and H/P/D across polls |
| Real data: available-wave sensitivity | How results differ when each wave uses all observed respondents |
Synthetic statements, participants and annotations are invented. The policy thresholds demonstrate software behavior; they are not a validated standard for judging deliberation. The real data supply surveys and group labels, so the report leaves transcript and briefing diagnostics unavailable.
Reproduce the reports
From the development repository, with deliberately installed:
This writes standalone HTML reports and audit R objects to
artifacts/gallery. The real-data import reads the frozen
polardata.tab and attitude-indices.tab
benchmarks from dp-data, under
evidence/benchmarks/. The bundled source manifest pins
their revision and hashes. Source files are hashed; exact duplicate
removal and endpoint rounding are logged. No participant data are
bundled with this package.
The documentation website serves only aggregate HTML reports. The saved R objects retain evidence and identifiers and belong in the research workspace.
Read a report in order
Start with the findings and plots, then check response and recording coverage. Use the diagnostic explorer to inspect an event, item, subgroup or method. Read its denominator and assumptions before interpreting the estimate. Policy verdicts follow the declared rules; missing inputs never create a PASS.
A large opinion change can coexist with poor process evidence, and almost no opinion change can coexist with thoughtful discussion. Read participation, argument consideration and learning together rather than treating movement as a success criterion. The design review explains the limits of each claim and the evidence needed to support stronger conclusions.