nx_gendiff.nx

nx_gendiff.nx -- the GENERATION DIFF that nx_plane_repair names and does not have. WHY. knowledge/store/debt- reads declared_qn=3696 beyond_qn=1: ONE key is reachable past the declared count, and that single key fails every nx_debt add/eat CLOSED, because add/eat re-seed the whole plane from the loaded buffer and committing would orphan it. nx_plane_repair refuses for exactly the right reason -- "reseeding would revive eaten rows and duplicate live ones -- that needs a generation diff, not a reconcile." Both organs failing closed is the system working. What is missing is the ability to SAY WHAT THE EXTRA ROW ACTUALLY IS. (STAR)A REFUSAL THAT NAMES THE MISSING CAPABILITY IS A SPECIFICATION -- BUILD THE THING IT NAMED, DO NOT WEAKEN THE REFUSAL. THE AMBIGUITY, STATED PRECISELY. A key past q:n is one of three things, needing opposite actions: SUPERSEDED byte-identical to a loaded row -> already represented, safe to leave behind REVIVED an id an earlier eat deliberately closed -> must NOT come back LOST-APPEND a DISTINCT row never counted -> real filed work; discarding it is silent data loss The estate's loader documents only the SUPERSEDED story because it was reasoned about for a plane that SHRANK. This plane GREW (3,621 -> 3,696 today), which makes LOST-APPEND the leading hypothesis and the exact case where reconciling q:n downward destroys evidence. (STAR)A STORE'S DOCUMENTED FAILURE MODE IS THE ONE ITS AUTHOR HAPPENED TO MEET; THE OPPOSITE DIRECTION GETS THE SAME SYMPTOM AND THE OPPOSITE REMEDY. THIS ORGAN WRITES NOTHING. It classifies and prints; deciding is a separate deliberate act. license_tier: ORIGINAL No hw writes. Read-only.