Every element of a good design exists to block a specific way of being wrong.
Controls block "it would have happened anyway". Your control group receives everything except the treatment. If your treatment involves a solvent, the control gets the solvent alone — otherwise you have tested the solvent, not the compound.
Replication blocks "it happened once by chance". Independent replicates mean genuinely separate experimental units, not three measurements of the same plant. Three trials is a common minimum; it is a floor, not a target.
Randomisation blocks systematic bias you did not think of. Assign units to groups with a random number generator, not by hand — "I put the healthy-looking ones in the treatment group" is unconscious and very common.
Blinding blocks your own expectations. If you are scoring anything subjective, have the samples relabelled by someone else so you do not know which group you are scoring.
Standardisation blocks confounds. Everything except the independent variable is held constant and written down: same light, same water volume, same time of day, same instrument.
Write the full procedure before you start, then have someone else read it and try to find the step where a difference could creep in that is not your variable. That hour is the highest-value hour in the project.