Improving breeding has been widely utilized in crop breeding and contributed to yield and quality improvement, yet few researches have been done to analyze genetic architecture underlying breeding improvement comprehensively. Here, we collected genotype and phenotype data of 99 cultivars from the complete pedigree including Huanghuazhan, an elite, high-quality, conventional indica rice that has been grown over 4.5 million hectares in southern China and from which more than 20 excellent cultivars have been derived. We identified 1,313 selective sweeps (SSWs) revealing four stage-specific selection patterns corresponding to improvement preference during 65 years, and 1113 conserved Huanghuazhan traceable blocks (cHTBs) introduced from different donors and conserved in >3 breeding generations were the core genomic regions for superior performance of Huanghuazhan. Based on 151 quantitative trait loci (QTLs) identified for 13 improved traits in the pedigree, we reproduced their improvement process in silico, highlighting improving breeding works well for traits controlled by major/major + minor effect QTLs, but was inefficient for traits controlled by QTLs with complex interactions or explaining low levels of phenotypic variation. These results indicate long-term breeding improvement is efficient to construct superior genetic architecture for elite performance, yet molecular breeding with designed genotype of QTLs can facilitate complex traits improvement.