SQLite Recovery Benchmark: harness, test data and results for eight forensic tools

Benchmark harness, test databases, answer keys and raw results of a comparison of eight SQLite recovery tools (FQLite, sqlite4n6, DC3 SQLite Dissect, bring2lite, SQLite GUI Analyzer, NFI xsqlite, sqlite_extract, SQLite .recover) on the SQLite Forensic Corpus (categories 0A-0E and the anti-forensic categories 11-19) and three false-positive scenarios after Lee et al. (2025).

Metrics: recall and precision of deleted-record recovery, recovery of live data in manipulated databases, false positives after B-tree rebalancing, drop-and-recreate and WAL with secure_delete, and robustness (crashes, time-outs, memory-limit violations). Each tool is run through an adapter that converts its output into a common format; all runs are limited to 180 s and 4 GB. The record contains the harness (scripts and adapters), the corpus databases, the ground-truth files, the normalised output and logs of every run, and the aggregated report. See README.md for contents, tool versions and reproduction steps.

Companion data set to the paper 'Recovering Deleted and Manipulated Records from SQLite Databases: A Reproducible Benchmark of Eight Forensic Tools'. Licences: benchmark code Apache-2.0, results and answer keys CC BY 4.0, SQLite Forensic Corpus databases public domain (CC0), see LICENSE-DATA.md.