For a guided tour, read Read a Keplar answer. This page is the lookup table.
Consensus levels
| Level | Meaning |
|---|---|
| Consensus | Most or all models answering held the same position |
| Partial | A majority position with at least one different view |
| None | Models split with no clear majority |
| Single | Only one model answered, so there is nothing to compare |
| Unscored | Agreement was not measured, for example for exact-check or small-talk answers |
Agreement is shown as counts ("4 of 5 models") rather than percentages. A missing model is not counted as disagreement. Internal thresholds of roughly 80 and 60 percent separate the levels. There is no confidence or accuracy score.
Sections
| Section | Contents | Appears |
|---|---|---|
| Consensus | The level and who held which position | Multi-model answers |
| Models consulted | Models that answered, any that failed or timed out, stand-ins | Always |
| Sources | Links models cited, labeled "Cited by <model>; Keplar didn't open or check" | When models cited any |
| Disagreements | The positions that differed and the reasoning for each | Only when there is more than one position |
| Verification | What the reviewer or judge examined | When a review ran |
| Reasoning summary | Short explanation of how the answer was formed | When provided |
Verification outcomes
| Outcome | Meaning |
|---|---|
| Held | The point stayed in the draft after review |
| Revised | The point was changed |
| Removed | The point was dropped as unsupported |
| Kept as dissent | The point was kept and attributed to a minority |
Issue kinds are reasoning problems, contradictions between responses, and unsupported claims. Verification reads the draft against the models' responses. It does not open sources.
Other fields you may see
- Thoroughness in effect: Automatic, Fastest, Balanced or Most thorough.
- Exact check marker for computed answers.
- Notes when an attachment could not be read or a model failed.
Related
- Read a Keplar answer: A tour of the six sections under every multi-model answer, what each one means, and what it does not mean.
- Agreement and the consensus level: How Keplar groups responses into positions, weights them by support, and turns that into a consensus level, and what the number does and does not mean.
- The verification review: What the verifier or judge actually does: it reads the draft against each model's response and flags reasoning gaps, contradictions and unsupported claims. It does not browse or open sources.