v2.79.2 - one rule, tested, in both places

Changes since v2.79.1

Write the disagreement rule once, and test it

The same decision existed in two branches a few lines apart - one comparing
memory against the stored maximum, one against what the numbers had just
read - and only the second ever ran. Fixing the first and releasing it
changed nothing, twice, while the log printed the old message unchanged.

One function now, called from both places, with a test that names the real
failure: a box reading 100 for a pool of 191 loses to memory; with no
structural lock neither fires; agreement is left alone.

Co-Authored-By: Claude Opus 5 noreply@anthropic.com

Full Changelog: v2.79.1...v2.79.2