- review mapping: CR-01 to Q1, WR-01 to Q2, WR-02 to Q3, WR-03 to Q4, each with its fix commit and closure state
- gate results, per-defect reverse-verification evidence and the re-recorded test:ci baseline
- escalations: the user-owned equipment.test.ts replay-stub drift blocks the CR-01 witness; WINDOWS.md intentionally untouched; ROADMAP.md/STATE.md left to the orchestrator
- clone() now binds each cloned modifier to the clone, records its modifierName, mirrors the source save flag and recomputes the attribute, matching addModifier's bookkeeping
- a cloned attribute can enumerate, index and persist its modifiers, and a cloned modifier's setValue notifies the clone
- add the bookkeeping and save-ability regression cases
- both multiplier checks tighten from < 1 to <= 1, matching warning 149's own text (must be greater than 1)
- an exact multiplier of 1 now warns 149 and falls back to the existing factor 2, so buffer expansion always terminates
- add the multiplier-of-one regression case (two code 149 warnings, 15 appended steps all read back in order)
- normalizeParam returns null for an unencodable param value, matching its own jsDoc and the documented discard semantics of normalizeParamList
- the command param count, the bytes actually written and the param cursor stay consistent so later commands keep their correct byte offsets
- add the discard-alignment regression case (count 0, later command bytes intact, read stream aligned)
- compareEquip clones each compared item's modifier before adding it to the comparison clone and removes the same clone objects afterwards
- the live attribute's already-bound modifiers never enter the clone and are never unbound
- add the equipped-item diff regression case (both directions, no warning 108, live bookkeeping untouched)
