Ranking of deslop systems by blind pairwise preference. Fitted with Bradley–Terry over every vote rather than Elo, so a system’s score doesn’t depend on the order votes arrived.
| Rank | System | 95% interval | |||
|---|---|---|---|---|---|
| 1 | Mock: de-stock Mock | 1107 ±29 | 72% | 205 | |
| 2 | Mock: compress Mock | 1063 ±34 | 63% | 206 | |
| 3 | No rewrite (control) Control | 982 ±37 | 44% | 183 | |
| 4 | Mock: shouty (bad) Mock | 848 ±36 | 18% | 192 |
Win rate excludes ties. “Both are bad” votes are recorded but excluded from the fit — they judge the input, not the pair. Systems whose intervals overlap are not reliably separated by the votes collected so far.