Optimal f finds the fraction of capital that maximises the geometric mean of a specific series of past trades. Size is derived by dividing that fraction of equity by the largest loss in the sample, then buying that many units.
It is kelly-criterion logic generalised to arbitrary trade outcomes rather than fixed odds, and its weakness follows directly: it is fitted to the worst loss you have already seen. One new trade that exceeds it makes the historical optimum retrospectively far too large. Tomorrow's biggest loss is usually bigger than today's.
Typical optimal f values on real systems land between 10% and 40% of equity, with implied drawdowns above 50%. Almost nobody trades it raw. Treat it as an upper bound that tells you where the cliff is, then trade a small fraction of it, the same way you would use fractional-kelly.
Related: fractional-kelly, kelly-criterion