Appendix — Change control
Summary
The appendix records what the two source briefs settle (GM1–GM8: the grant discovered not authored, authority not authorisation, the library/instance split, reality before risks, the three-term comparison, memory as a time axis, adopt-Cedar-and-graphs-build-only-the-mandate, and the naming checks) and what the pack's own construction added (GM9–GM10: the first library entry refusing to measure itself, and node n3 as the thesis caught live). The decisions register carries twenty-nine entries. GM11-GM14 record build-order step 1 built and tested, and GM15-GM16 record the library's second entry — measured inside a CI runner, joining the first at node n3 — together with the two defects it found: a tier mislabelled because it was decided in isolation, and the pre-push hook not travelling with a clone: the acceptance test passing, the control refusing the release that carried it, the measurement catching its own tier change, and the standing limitation that the authority is a fixture while the enforcement is not — who measures a library entry and with what authority, how often it is re-measured, whether the instance records which library version it referenced, the history-window field, where a built pack lives, whether the blind-spot number is published per agent, whether a hook can trust the mandate it reads, and the corpus version on adoption.
Key concepts
- Supersede, never rewrite — the same discipline the sibling register uses
- GM9 — the refusal as evidence — a boundary control caught working on the measuring agent
- The eight open decisions — measurement authority, cadence, version-pinning, and where the pack lives
Key ideas
- The delta is computed not stored, carried from the register's C7.
- A future correction to a shared decision lands in both registers with a cross-reference.
- The corpus version is assigned on adoption — this is a first pass, for review.