Analysis and code archive: Understanding modality-wise completeness estimation in language-dominated multimodal sentiment analysis under random input missingness

Machine-readable analysis and code archive for a PeerJ Computer Science submission. The archive contains complete reconstructed code trees for the modified and original LNLN implementations, raw evaluation CSVs for CMU-MOSI and CMU-MOSEI, per-test paired-comparison statistics with Holm corrections, completeness-estimator calibration audits, evaluation-mask sensitivity analyses, efficiency benchmark outputs, sensitivity power analysis, publication figures with provenance, checkpoint manifests, analysis scripts, training/evaluation logs, configuration files, and dataset/checkpoint manifests. Checkpoint binaries are not included because of their size; their SHA-256 hashes, sizes, best epochs, and validation MAE are recorded so that retrained checkpoints can be verified. Third-party dataset feature files are not redistributed; their official sources, sizes, and checksums are recorded.