Checking that our own footnotes point at something real

VEILOS publishes a page describing how it keeps your data consistent when several things happen at once. For each of the four mechanisms it lists, that page names the exact file the code lives in, the exact functions inside it, and the exact test that proves the mechanism works — and then calls the list complete. This release found that nothing had ever opened any of those files. The only check was that each name began with the right few letters, which means the letters on their own would have passed, and a test deleted months ago would have kept passing forever. Every one of those references is now followed: the file has to exist, the test has to exist, and the function named has to genuinely appear in the code rather than merely be mentioned in a comment beside it. All twelve references were checked and all twelve are real today — nothing on that page was false, and this is a lock fitted before anything was stolen. The word “complete” has also gone. The list called itself complete over a count of four, and four was simply how many were written down; nobody had ever asked how many places in the system actually save data. Counted properly: one hundred and seventy-four places across seventy-four files. The page now says that four of those files are covered here and seventy are not — carefully worded as a question we have not asked rather than an accusation, because most of them almost certainly rely on a mechanism already listed and we would rather admit we cannot yet prove it than imply either answer. Finally, the same check was turned on last release's own work and found the identical mistake hiding there, in the one line whose entire job was to tell you where to look. That was fixed in the same breath.