canvas-humanizer-loop

Nested-loop wrapper around canvas-humanizer that runs the round-trip humanizer, audits its output for meaning/structure/rubric damage via 3 parallel sub-agents with majority vote, dispatches per-segment 1-in-1-out rewrites for damaged segments, then re-humanizes (with already-converged segments locked) — up to max_iter=3 with 3 layered convergence guards (MUST_FIX==0, per-segment verdict monotonicity,…