The marker vocabulary had six entries - REQ, PREF, PROVEN, ASSUMED, DEFERRED, INSTANCE. They could say how strong a requirement was and where it came from. None could say it was not being met. That gap is why five unmet REQs in this document went unrecorded for weeks: there was no notation to write them in, so nobody wrote them.
DIVERGENCE is added as the seventh, with the rule attached. A REQ that stops being met does not become a PREF and is not rewritten to describe what was built. It stands, the gap is recorded, the correction is owed by the thing. A specification that agrees with whatever exists specifies nothing. No requirement in this revision was lowered.
DIV-001 at section 14 item 11: the AGPL section 13 source link is a numbered constraint the application code will follow, and it is not met on a public deployment. DIV-002 at section 9 twice and section 8.3: the FastAPI control plane, the worker unit and the SQLite queue do not exist, and five declared dependencies are imported by nothing. DIV-003 at sections 1.3 and 5.2: the instance is publicly reachable against three REQs. DIV-004 at section 11: eight declared keys are read by nothing.
Section 11 also gains MECHCOMP_MAX_EXPORT_MM, read by the code since cdde394 and declared nowhere. That is the divergence running the other way and the harder one to notice, because a missing key looks like nothing at all.
Section 5.2 records that the divergence may be a misfiling rather than a violation. WORK-ORDER-004 section 3 says dev abbreviates Mechanical Compiler Developers, that the name is production, and that it appears on printed material. If that reading holds, 5.2 never applied to this instance and 5.3 does. Written as a decision, not a correction.
Section 1.1 gets a note that it was right about CadQuery all along. DIV-005 was opened against this paragraph rather than against HANDOFF section 5, which had it backwards for three weeks.
Section 15 gate 3 now states its status: dependencies, toolchain image and scaffolding all met, worker unit not met, gate does not pass. Section 19 records the port complete since 20 AUG, retained because the rejected cases being part of the contract is why the oracle means anything.
STAGING-STATE specification reference updated from revision 5 to 5.4.
Applied by anchored patcher. The first attempt failed on one anchor - the env block comment column was 35, not 39 - and nothing was written, including the thirteen correct patches. Suite 643 passed, oracle intact. Documentation only.
TheRON