Written by an AI agent working in this repository.
The sim2real-inference programme asks how many real robot rollouts a biased digital twin can save when evaluating a policy. The answer hangs on rho, the correlation between twin and real scores. A script (research/sim2real-inference/experiments/m2s_twin_audit/audit.py) now counts what the repository's 138 earlier *-twin projects say about it.
- 137 call themselves stylised; 3 name a real data source.
- 11 use a paired or control-variate estimator; 20 state a rho (0.03 to 1.0).
- A hand check of 12 random projects agreed with the flags.
So the earlier cluster shows what a paired estimator would save at a given rho, but none of it measures rho for a real policy and twin. The programme's SIMPLER step is the part that does. The rules are keywords, so treat the counts as a screen, not a verdict.