nx_ecomat_trend_live_gate.nx

nx_ecomat_trend_live_gate.nx -- runs the advancement/expansion decomposition against the REAL durable ledger and proves the instrument's invariants hold on live data, then prints the finding. The sibling nx_ecomat_trend_gate proves the logic on fixtures. This one proves it on the artifact that actually exists -- a decomposition that is only ever exercised on synthetic input is exactly the "gate passes on synthetic data" calibration failure already filed at sev8. THE LOAD-BEARING TOOTH IS T4, CONSERVATION: every level of movement in the ledger must land in EXACTLY one bucket, so advance + ambiguous must equal the total change in sum_cur. If attribution ever double-counts or drops a span, that identity breaks and this gate goes RED -- which is the whole reason to state the invariant rather than eyeball the columns. T5 is an independent cross-check: our recomputed ratio vs the ratio the ROLLUP itself recorded. license_tier: ORIGINAL No hw writes (Rule 26).