Pipeline Yield — from remote reads to an annotated, benchmarked call set

Every module was verified against its own real output before the next one ran. This is that verification trail: read-level QC through alignment and deduplication, then variant-level yield through calling, concordance, and annotation — all for the same CAPN3 (chr15) + DMD (chrX) padded regions, region-extracted directly from GIAB's public HG002 BAM.

Read-level (EXTRACT_REGION → FASTQ_QC → TRIM_READS → BWA_ALIGN → SORT_MARKDUP)

336,549 read pairs extracted directly over HTTP range requests — no full-genome download — then fastp-trimmed to 325,255 pairs before alignment (see FASTQ QC: Raw vs Trimmed). Mapping/dedup stats below are for the trimmed reads BWA_ALIGN and SORT_MARKDUP actually process.

Variant-level (GATK_CALL / DEEPVARIANT_CALL → CROSS_CHECK_VCFS → annotation)