We have analysed 5-epoch GOODS HST-ACS B , V , I _ { 775 } and z datasets ( V1.0 release ) , in conjunction with existing VLT-ISAAC imaging in the J , H and K _ { s } bands , to derive estimated redshifts for the sample of 198 Extremely Red Galaxies ( ERGs ) with K _ { s } < 22 ( Vega ) and ( I _ { 775 } - K _ { s } ) > 3.92 selected by Roche et al .
( 2003 ) from 50.4 arcmin ^ { 2 } of the GOODS/CDFS field .
We find that , at this depth , the ERG population spans the redshift range 0.5 < z _ { phot } < 4.75 and over two decades in mass ( \sim 3 \times 10 ^ { 9 } M _ { \odot } to \sim 3 \times 10 ^ { 11 } M _ { \odot } ) .
Our results show that the dust-corrected red envelope of galaxy evolution is well modelled by a starburst at redshift z _ { f } = 5 followed thereafter by passive evolution .
We explore the evolution of the ERG luminosity function ( LF ) from redshifts \langle z _ { phot } \rangle = 1.0 to \langle z _ { phot } \rangle = 2.5 and compare it with the global K _ { s } -band LF at redshifts 1 < z _ { phot } < 2 .
We find that the bright end of the ERG LF does not decrease from redshifts \langle z _ { phot } \rangle = 2.0 to \langle z _ { phot } \rangle = 2.5 and we connect this fact with the presence of progenitors of the local L > L ^ { \ast } 
population at redshifts z _ { phot } > 2 .
We determine lower limits of \rho _ { c } = ( 6.1 \pm 1.9 ) \times 10 ^ { -5 } Mpc ^ { -3 } and \rho _ { c } = ( 2.1 \pm 1.1 ) \times 10 ^ { -5 } Mpc ^ { -3 } on the comoving densities of progenitors of local massive galaxies already assembled at redshifts \langle z _ { phot } \rangle = 2.5 and \langle z _ { phot } \rangle = 3.5 , respectively .
We have investigated the existence of high-redshift Lyman break galaxies massive enough to be included in this ERG sample .
Out of an initial list of 12 potential very high redshift candidates , we have identified 2 ERGs which have a high probability of lying at z _ { phot } > 4 .
We discuss the advantages of multi-colour to single-colour selection techniques in obtaining reliable lists of very high-redshift candidate sources , and present revised lower redshift estimates for sources previously claimed as potential z > 5 dropouts in recent studies .