Emailing the privacy address now guarantees someone follows upEmailing the privacy address now guarantees someone follows up
September 24, 2026 · privacy@jiolab.dev now survives the purge and files a task for every message it gets, after catching a Cloudflare quirk that had erased the address on one site entirely.Emailing privacy@jiolab.dev now guarantees a response: the message is kept and creates a task automatically, after catching a Cloudflare bug that had erased the address.
last updated
privacy@jiolab.dev is now a working deletion-request address end to end, not just a listed contact. Mail-catcher, my inbox service, purges mail after seven days and caps how much any one address can hold. Both are exactly wrong for a privacy request, which needs to survive and needs someone told about it. Today privacy@ became the one address exempt from both. Reading that inbox now also requires an admin session, since a published address carries no credential of its own and anyone can write to it. Every message that lands there files itself as a task automatically, with tests covering it, so a request sent into a published mailbox can no longer just sit there unread.
The catch came from checking the live page rather than trusting the build. Both this site's policy pages and dice-lock's privacy page name privacy@jiolab.dev as the deletion contact. On both, Cloudflare's Email Address Obfuscation was quietly rewriting the mailto link into a meaningless placeholder for any visitor without JavaScript, which ate the address entirely on dice-lock's first attempt to publish it. Both sites now opt that one link out of obfuscation explicitly.
Elsewhere, a different kind of check paid off. The keep-alive job that stops my Claude CLI credential from expiring runs on a timer and had always reported success. Reading six real renewal cycles in its logs showed the first spawn of every cycle authenticates fine but renews nothing (the CLI's own refresh window is narrower than the thirty-minute gap between wakeups), and only the retry twenty minutes later does the actual work. That's two authentication calls roughly every eight hours, not the one that three separate write-ups had claimed. The job now reports not-refreshed instead of misreporting renewed.
Dice-lock's server stopped hand-copying its own route list, too. Routes used to live wherever the dispatcher, the error responses, and the README each happened to repeat them, and they had already drifted: one route was missing from the README for a commit after it shipped, and nothing caught it. They are now generated from a single table the dispatcher walks. Alongside that, around 88 places that threw a bare string as an error became coded classes carrying a domain and an HTTP status. That fixed a real bug: four different causes of a 401 on /account (including an outage on the outside login service dice-lock depends on) were all being read by the client as "sign out," so an outage that was nobody's fault was logging players out of a game they hadn't done anything wrong in.
privacy@jiolab.dev is now a real, working way to ask me to delete your data, not just an address that looks official. The system I use to manage incoming mail normally deletes messages after seven days and limits how many any one address can hold, which would have been exactly the wrong behavior for a deletion request, since that message needs to survive and someone needs to be told about it. Today, privacy@ became the one address exempted from both rules. Reading what arrives there now also requires logging into the site's admin area, since a public email address has no password of its own and anyone can write to it. Every message that arrives there now automatically creates a task on my to-do list, with tests to prove it happens every time, so a request sent to a published address can no longer just sit there unread.
The catch behind that fix came from checking what a visitor actually sees, not just what the code says. Both this website's policy pages and the dice game's privacy page name privacy@jiolab.dev as where to send a deletion request. On both, a Cloudflare feature meant to stop spam programs from harvesting email addresses was quietly rewriting that link into a meaningless placeholder for anyone browsing without JavaScript turned on, which wiped out the address entirely the first time the dice game tried to publish it. Both sites now tell Cloudflare to leave that one link alone.
Elsewhere, a different kind of checking paid off. A background job keeps the login credential I use for automated coding sessions from expiring, running on a timer, and it had always reported success. Looking at six real cycles of its logs showed that the first attempt in every cycle logs in fine but renews nothing, because the login tool's own refresh window is narrower than the half-hour gap between wakeups, and only the second attempt, twenty minutes later, actually renews anything. That's two login calls roughly every eight hours, not the one that three different write-ups had claimed. The job now reports accurately that nothing was renewed on that first attempt, instead of claiming success it hadn't earned.
The dice puzzle game also stopped keeping its list of web addresses in several places by hand. Those addresses used to live wherever the code that answers requests, the error messages, and the read-me file each happened to repeat them, and they had already drifted apart: one address was missing from the read-me for a commit after it shipped, and nothing caught it. They are now generated from a single list that everything else reads from. At the same time, around 88 places in that game's code that just threw a plain error message got turned into proper labeled errors that say which part of the system failed and how badly. That fixed a real bug: four unrelated causes of one login error, including an outage on the outside login service the game depends on, were all being read by the game as "sign the player out," so an outage that had nothing to do with the player was logging them out of a game they hadn't done anything wrong in.