On Sat, 18 Jul 2026 at 03:30, Robert Elz via tz <[email protected]> wrote: > | This is an interesting and reasonably well researched article. > > Not really, the author was obviously fed a whole bunch of facts > (or used an AI to fetch them perhaps), but aside from: > > | However, there is a badly misstated paragraph: > > which was indeed a glaring error, there's lots more that is wrong > with it. >
I noticed the poor quality of the article, too. In particular, these two statements stood out as most jarring to me: "Any system that uses elapsed-time calculations to rate-limit operations, generate sequence numbers, or verify audit logs built on the assumption that time only moves in one direction — and only stops momentarily, never jumps." This feels backwards, at best. Any system that can handle the positive leap seconds we've already had can, in fact, deal with the fact that "time" either slows, stops, or repeats, depending on the implementation. The negative leap second is arguably the easier case, because time *does* only end up moving in one direction — the discontinuity is the hard part, yes, but no harder than the positive ones. Either speed up a bit or skip, but time is always moving forward. "At the rate Earth's rotation currently varies, a one-hour threshold would not be reached for somewhere between 50 and 100 years, effectively making both positive and negative leap seconds a concern for the distant future." We managed with only 27 in the last ~half-century, and now 0 in a decade, but somehow there's an imminent threat that we might need another 3600 so quickly that we'd literally run out of months to put them in one by one? The math isn't mathing here. It's not even remotely close. On Sat, 18 Jul 2026 at 07:13, Brooks Harris via tz <[email protected]> wrote: > I have experimentally implemented hypothetical negative leap-seconds. It's > a bit tricky in some respects. But one thing I've observed is that the > omission of YMD 23:59:59 leads to a natural roll-over to the next day. > (23:59:58 => 00:00:00). It's possibly less disruptive than positive > leap-seconds. It's possible many systems would just roll through that > without serious effects. The "Fear of Negative Leap-seconds" may be > exaggerated. > Yeah, while arguments for raising the DUT threshold beyond 0.9 seconds exist that could be reasonable, I could see that for a threshold of, like, 2 or maybe 5 seconds *if and when it's needed —* which it's simply not at this point. If there's concern that there hasn't been enough testing on negative leap seconds, imagine folks generations from now grappling with a leap minute that no one's even *built* anything for. And if Spanish culture has kept ticking along despite being "numerically misaligned" with the sun for nearly a century, imagine getting the entire world to agree on a leap *hour* after cultural norms have spent a millennium drifting and adapting. It would make present-day DST arguments look like child's play. As such, I've long held that the better solution to grappling with the realities of these perturbations is to better *routinize* how we reconcile them — which the existing system, along with various industries' innovations like smears, actually handles pretty darn well. We've known about the possibility for a while, the possibility has been more real for several years now… and actually, plenty of people *have* tested it and continue to do so. But what do I know? </soapbox> Anyway, the phenomenon of an article based on a kernel of "real technical news" being repackaged as a "common interest" story and supplementing it with shoddy background research is not at all new… but with LLMs now capable of doing bad research quicker than ever, I fear we'll soon see a lot more stuff like this. -- Tim Parenti
