A 13-month calendar and decimal world time, specified and implemented
Date arithmetic is needlessly hard, and almost all of the difficulty is inherited rather than essential. Months have four different lengths for reasons that stopped mattering two thousand years ago. The same weekday falls on a different date every year, so no schedule repeats cleanly. Time zones and daylight saving turn a global meeting into a lookup problem, and the BC/AD split puts a discontinuity in the middle of recorded history. Every one of these is a decision that could be made differently, and software pays for all of them.
Thirteen months of twenty-eight days give 364 days, and every month begins on a Monday. The consequence is that any date has a fixed weekday forever — the first of any month is always a Monday, scheduling repeats without a lookup table, and month-length branching disappears from date code entirely. It also repairs the etymology: September becomes the seventh month again.
364 days leaves a remainder against the solar year. Rather than lengthening a month and breaking the symmetry, GHS inserts a full seven-day intercalary week every five to six years. The week structure survives the correction, which is the property that makes the whole calendar predictable. The year starts at the northern spring equinox, which at least has an astronomical basis — 1 January does not.
The day is divided into 1,000 decimal beats with no time zones and no daylight saving. @685 is the same instant everywhere on Earth. This trades local intuition (noon is no longer a fixed number) for global unambiguity, which is the correct trade for coordinating anything across borders.
Adding 10,000 years to the Gregorian year gives a single continuous count across roughly twelve thousand years of civilisation, with no sign flip and no era boundary in the middle of the timeline. A full GHS timestamp reads 12026.01.15 @685.
The repository carries the specification plus implementations in TypeScript, Kotlin, Swift, Dart and C++. A time standard that exists only as a document is a blog post; the point of shipping conversion code for several platforms is that the edge cases — Aurora Week, equinox alignment, round-tripping to Gregorian — have to be resolved rather than waved at.
Calendar reform has been proposed roughly once a generation since the 1800s and has never survived contact with the installed base. I am not expecting this one to either, and the project does not pretend otherwise. What makes it worth building is the exercise: writing a complete time specification and then implementing it five times forces you to confront every assumption that ordinary date libraries paper over — and it produces a working reference for anyone who wants to argue about the design with code rather than opinions.