Checking the front door, instead of remembering that we checked it

A while ago someone confirmed that mail sent to this project's contact address actually arrives. That confirmation was then written down by hand and never checked again — and four separate release checks have been reading it ever since as though it were current. It was thirty-five days old. Now a small tool checks what can actually be checked, every time it runs. It asks the public internet whether the mail records for the domain are in place — all four are — and it asks the mail provider directly whether the domain and the sending address are verified, which they are. Those answers come from outside this project, which is the point: a claim about the world should be answered by the world. What it deliberately does NOT do is guess about the rest. Proving a message actually arrived means sending one, and proving the address can reply as itself means a person replying and someone else reading the header. Neither is something a script can invent, so both are left explicitly blank rather than filled in with something plausible. The two release checks that depend on them stay honestly unsatisfied. The interesting part was a mistake caught just before it shipped. The first version threw away the old delivery confirmation and stamped today's date on the record. Both were wrong, in opposite directions. Throwing it away would have turned "we confirmed this, a while ago" into "we never confirmed this" — destroying a real observation. Stamping today's date would have been worse: that date is exactly what the staleness check reads, so a thirty-five-day-old confirmation would suddenly have looked fresh, purely because a different thing had been checked today. That is the precise dishonesty the whole exercise was meant to remove, arrived at by accident inside the tool built to prevent it. It now keeps two dates, because there are two observations, and the check that watches one is never handed the other.