There was an RC discussion as to what the signals target represents.
This notebook decode_signals_target explores the question. The official target and the rank of 2day-6day (per documentation) returns are 0.71 correlated.
The computed field matches 63% of the official targets exactly, and is only one category away (e.g. 0.25 vs 0.50) for 35% of the cases. Only 1.5% are two or more categories away.
I am chalking up the one-category difference to: i) mostly not having as many tickers from yahoo as the numerai internal data, and ii) early-exit criteria in the target for profit-taking and stop-loss. The final 1.5% is likely some Michael Oliver secret sauce.
Conclusion: key component of the official target is the binned ranked 2day-6day return, as computed in this notebook.