# Supplementary analysis: self-preference adjusted for judge harshness

Added Thursday 1 October 2026, after the results were published internally and following an independent review of the write-up. This is an addition to the record. The pre-registered method in `briefs-method-and-judging.md` and the original self-preference figures in `results.md` are unchanged.

## Why this was added

The original measure (set before any scores existed) compares the score a judge gave its own blind entry with the mean of the other seven judges' scores for that entry. That gap also captures whether a judge is generally harsh or generous. A judge that marks everyone down will tend to show a small or negative own-entry gap even if it rates its own work relatively well.

## The adjustment

For each judge: (mean own-entry gap across the five events) minus (mean gap on the other 35 entries it judged, each measured against the other seven judges' mean for that entry). Every row is in `supplementary-self-preference-adjusted-2026-10-01.csv`.

| Judge | Own-entry gap (original measure) | Gap on others' entries (general harshness) | Adjusted |
|---|---|---|---|
| GPT | +0.63 | -1.02 | +1.65 |
| Claude | -0.09 | -1.03 | +0.94 |
| Copilot | +1.44 | +0.50 | +0.93 |
| Gemini | +0.83 | +0.28 | +0.55 |
| Kimi | +0.40 | +0.14 | +0.26 |
| GLM | +0.45 | +0.20 | +0.25 |
| Siri | +0.47 | +0.23 | +0.24 |
| DeepSeek | -0.15 | +0.13 | -0.28 |

## Limits

- One own entry per judge per event, five in all, so each adjusted figure rests on five data points. Treat it as a pointer worth investigating, not a verdict.
- It cannot show that any judge recognised its own entry, or favoured it deliberately. Entries were blind; a preference for one's own house style would produce the same pattern.
- Judges may be harsh in different ways on different events (for example, one weak entry pulls a mean down), and a single average does not capture that.
- This was not part of the method set before the run. It is reported because the original measure, on its own, flattered the harshest judges.