D329 - The one table left hand-copied, and the rule that seemed to forbid generating it
decided · 2026-08-27 · found because a generated number moved and a typed one did not
docs/PROJECT_STATUS.md carried a per-title table -
reach, imports, calls, standing, where each ends - typed by hand. It
said PPSA02664 reached 23 imports and ended at
image+0xafc959. The records in compat/ said 25
and image+0xafcc08, and had done since a run earlier the
same day. Four days of drift in the one table a reader looks at
first.
D240 is why it was typed: a generated block may hold only numbers the tool can recompute anywhere, because the corpus is not tracked and a block claiming "6 of 6 titles execute guest code" would fail for every contributor who owns no titles. That reasoning is about needing a run.
compat/ is committed. Reading it recomputes from what
the repository ships, works in CI, and works for somebody with no titles
at all - so D240 permits this, and the table is generated now. Worth
writing down precisely because it looks like a contradiction:
the rule is not "nothing about titles", it is "nothing that needs
one".
The honest slot only. [experiment] is a
real result about a different question, and one table holding both would
be the propped-up number wearing the honest one's clothes. Titles with
an experiment are named underneath, saying what they reached and how
many functions were answered by name (D312).
The small thing that would have made it read as a regression
The first generated table printed 98957030 where the
typed one had 98,957,030. Nobody would call that a bug and
everybody would notice it. A generated artefact replacing a hand-written
one has to be better in every respect a reader can see, or the
generation gets blamed for the loss and reverted - and the drift comes
back with it.