D316 - The decision number was allocated by a convention that races

decided · 2026-08-27 · after colliding twice in one afternoon

A decision is cited from source by number, so two entries sharing one makes every citation of it ambiguous - a reader following (D313) lands on two decisions and cannot tell which the code meant.

The convention was read the highest number, add one. That races, and not through carelessness: more than one session appends to this file, each reads the last number, and neither sees the other's unlanded writes. It happened twice in an afternoon, both times because a number was chosen, several minutes were spent writing the body, and by the time the entry landed somebody else had taken it.

The window is the bug, so ./orbistoun.sh decide "<title>" closes it. The number is claimed the instant it is chosen - under a mkdir lock, atomic everywhere this runs - and a reservation is appended before a word of the body exists. Whatever a second session does next, it reads a higher number.

The gate already caught duplicates and that stays: it is the backstop for anyone who edits the file by hand, which is still allowed and always will be. What changes is that catching one is no longer the only mechanism, and by the time it fires the damage is already in the source comments citing it.

A reservation nobody spends is now refused too. A claimed-and-abandoned number reads in the log like a recorded decision, and leaving one behind is the easy failure - a session ends, a context is lost. The guard was watched refusing one before this entry replaced it.

The general shape is the one this log keeps arriving at from different directions: a rule held by a person is a rule that holds until two people apply it at once. The fix is almost never to be more careful.