Reproducibility package: registry-required corroborating lines and reference-limit discordance for LIBS element detection in the ChemCam archive

Reproducibility package (version 2.2) for a methods study of automated element-detection scoring on the ChemCam LIBS archive (Mars Science Laboratory). Contents: the frozen candidate registry with registry-required corroborating lines per element (registry.json), the pinned NIST ASD line export used by the cascade (nist_lines.csv; ASD v5.12, queried 2026-07-05 via astroquery over 240-906 nm), the 2,053-cell legacy labeled evaluation panel (labeled_panel.csv), the full 4,464-cell score table with score-blind inclusion labels covering every scored cell (full_score_table_4464.csv; labels present-reference / below-external-limit / truth-unknown / indeterminate / unlabeled-no-reference), the frozen reference-composition table with per-value provenance (ccct_certificates.json), the product manifest of all 372 standard observations with per-product shot counts (product_manifest.csv), the 752-observation SuperCam seed manifest with exact PDS product URLs (supercam_seed_manifest.csv), the complete detector and threshold specification (threshold_method.md), the full analysis code (code/: pipeline stages s0-s5, reanalysis scripts r1-r11, the w5 verification script, and the round-3 scripts r12-r14 with their JSON outputs: score-blind labeling, shifted-registry/decoy chance-match controls, common-universe transfer comparison), SHA-256 checksums of all files (CHECKSUMS.sha256), and the manifest documenting software versions and random seeds (MANIFEST.md). The panel contains no certified true zeros: below-external-limit cells are cells whose certificate concentration sits below a published method-specific ChemCam reference limit, and reported call rates on that panel are reference-limit discordance rates, not false-positive rates against certified blanks. Threshold selection, cluster-aware (observation-level and class-level) bootstrap inference, and the dust-gate sensitivity analysis, and the round-3 chance-match controls are specified in threshold_method.md and implemented in the scripts referenced by MANIFEST.md. Version update: hardcoded local filesystem paths in the analysis code were sanitized to relative / ${DATA_DIR} form; no scientific content, number, or result changed.