Artifact for "The Shelf Life of Verification Evidence: A Pre-Registered Six-Year Study of Constant-Time Verdicts in Three Cryptographic Libraries"

Complete artifact for a pre-registered longitudinal study of constant-time (CT) verification verdicts in libsodium, Mbed TLS and PQClean across 26 quarters (2020Q1–2026Q2). Contents. The hash-committed pre-registration and its SHA-256; era-appropriate Dockerfiles for four Debian tiers (buster, bullseye, bookworm, trixie), each with an identical pinned TIMECOP/Valgrind 3.23.0 oracle; all harness generators and cohort files; the orchestration driver and the DAG-correct git bisect run bisection driver; resolved quarterly commit tables; and every raw verdict cell (377 CSVs, 2120 Valgrind logs). Headline numbers. 2236 (function, quarter, compiler) cells, 2014 testable, 43 subjects, one host. One admissible verdict transition, one repair, six identity deaths, two vacuous cells. Verification. verify_paper_numbers.py re-derives every number in the paper from the raw CSVs and asserts the vacuity guard is present in all three runners. full_study/test_analyze.py carries estimator regression tests. shasum -a 256 -c PREREGISTRATION.sha256 verifies the pre-registration against its published hash. Note on the corrected instrument. An earlier version of the runner captured each harness's exit status and never read it, so a harness that failed during setup was scored CT_PASS on zero executed code — a vacuous pass. All three arms now record exit status and emit a distinct VACUOUS state. The full sweep was re-run under the corrected instrument; results from before the correction are retained under full_study/<lib>/results_prevacuity/ for diffing.