Supplementary repository: code, configs, and evaluation artifacts for 'Open-World Source-Free Adaptation for Underwater Acoustic Target Recognition under Naturally Contaminated Target Streams: Adaptation Survives, Post-hoc Rejection Does Not' (to appear as Preprints 2026)

Supplementary repository accompanying the preprint 'Open-World Source-Free Adaptation for Underwater Acoustic Target Recognition under Naturally Contaminated Target Streams: Adaptation Survives, Post-hoc Rejection Does Not' (to appear as Preprints 2026; preprint v1.2, September 2026). Fulfills the Data Availability statement: upon preprint posting, this archival repository releases the code, configuration files, split identifiers where permitted, trained-model metadata, and derived evaluation artifacts. Contents: all 17 analysis/training/evaluation scripts; per-run configs, histories, and summaries (3 directions x 5 methods x 3 seeds); aggregate metric JSON and per-seed CSVs; unknown-subgroup rescoring tables and cached per-segment logits; threshold-transfer cell data; feature-space scoring results with aggregate source-stage statistics; calibration/bootstrap/sklearn cross-check tables; FPR@80/90/95TPR operating-point sensitivity tables; figure sources; sha256 manifest. No raw audio is redistributed; model checkpoints and third-party cached embeddings are excluded per the underlying dataset licenses and are exactly reproducible with the deposited scripts. VTUAD is available under concept DOI 10.5281/zenodo.22274549; Oceanship/QiandaoEar22/ShipsEar from their respective providers under their stated licenses.