A birth-time calculator that has to know about China's forgotten 1986 summer time

The general shape of this bug has nothing to do with astrology. It is treating a zone's present offset as its offset at some arbitrary past instant, which is what almost every wall-clock-plus-zone record in production quietly assumes.

Longitude is the part that never turns off. At four minutes per degree, China's single clock across five geographic zones puts a clock in Urumqi roughly two hours ahead of the sun, according to the author's writeup on dev.to ^[6]. Set that against twelve two-hour branches ^[1] and it is exactly one branch, for every birth, all year ^[1]. Vancouver's three degrees west of its standard meridian is quieter at 12 minutes ^[7], but it is there in January as much as in July.

The July Vancouver case stacks up to about 78 minutes of correction ^[8], which is roughly 65 percent of a branch ^[3]. Take out the 12 minutes of longitude and 66 minutes are left for summer time plus the equation of time ^[2], so the dominant term is the one that needs history rather than geometry. The equation of time itself, worth up to plus or minus 16 minutes over the year, is the small correction here, and the NOAA approximation the author uses is accurate to under a minute ^[4].

Pulling a historical offset out of the runtime is a tidy trick: Intl.DateTimeFormat will format a UTC instant in any zone, and reassembling those formatted parts as though they were UTC and differencing gives you the offset in minutes ^[10]. The flipside of shipping no data file is that a 1988 answer becomes a property of whichever ICU build your base image happened to carry.

The platform gaps are where the workarounds worth stealing sit. JavaScript exposes no isdst, so the code samples the zone offset on Jan 1, Jul 1 and the birth instant, taking the minimum as standard time on the reasoning that DST only ever moves clocks forward ^[11]. That third sample is not decoration. Morocco runs negative DST during Ramadan, and without it the heuristic invents a 60 minute shift that never happened ^[12]. Separately, computing local mean solar time as longitude times four minutes relative to Greenwich lands the Chatham Islands, at 176.5W on UTC+12:45, a full day out, straight into the day pillar the whole reading hangs on ^[13]. Normalising longitude to within 180 degrees of the zone's own standard meridian rather than Greenwich is the correction ^[14].

Going the other way, from wall time to an instant, is circular: you need the offset to find the instant and the instant to find the offset. The author reports two fixed-point iterations converge everywhere except inside the one-hour DST gap, where no exact answer exists to converge on ^[15].