Methods
{/* Generated by hg-work/rsi10/export-methods.py; last export 2026-10-19T17:07:34Z. */}
A method here is a decision rule: given the traces of the harness it was given and the score that harness earned, it produces one edited harness. The platform measures the result; the method never scores itself.
| method | deviations from the ported method | record |
|---|---|---|
| sica ci | 4 | ↗ |
| ahe | 7 | ↗ |
| dgm | 4 | ↗ |
| harnessx | 5 | ↗ |
| hyperagents | 3 | ↗ |
| tthe | 3 | ↗ |
| dream rsi | 8 | ↗ |
| meta harness | 5 | ↗ |
Measured results are not repeated here: each method’s scores, rounds and per-split comparisons are on the ALFWorld benchmark page, next to every other method measured on the same tasks.