boltzgen run input_spec.yaml --steps filtering --config filtering budget=60 alpha=0.0

. Filtering: each design is evaluated against mandatory thresholds. 2. Ranking: for every metric we compute its rank, then scale it by the metric s inverse-importanceweight. Designs with fewer passed filters are automatically penalised because the ranking key is the pair(num_filters_passed, metric). The overall quality score is the worst (maximum) of these scaledranks.The 10 best designs form the Top set. 3. Diversity: a lazy-greedy algorithm selects 2 designs that jointly maximise quality andminimise sequence similarity (sequence-identity distance). The trade-off is controlled by = 0.001:= 0 100 % quality focus (same as Top set)= 1 100 % diversity focus (ignores quality)Quality composite of metrics such as iPTM, salt-bridges, SASA, etc. Each metric has an "inverse importance" weight (see "Sorting Criteria" table). A larger weight divides the rank by a bigger number and therefore down-weights that metric. Diversity 1 sequence identity between designs.We use = 0.001, meaning 100 % emphasis on quality, 0 % on diversity