An ECO changes the netlist, so the previous equivalence result no longer applies. The patched netlist is compared against the RTL again with a miter before signoff.
The comparison covers the whole design, not just the edited cone. A wrong cone selection is itself a bug the full check catches.
Small RTL changes can still move area and delay. Guardbands below are stated EXAMPLE values, not tool defaults.
EXAMPLE guardbands (not tool defaults): area may grow at most 2%, delay may grow at most 1%
baseline (stated EXAMPLE values): 100 area units, 10.00 ns output arrival
after ECO: 101 area units, 10.05 ns arrival
area change = 101 - 100 = 1 of 100 = 1%, within the 2% guardband
delay change = 10.05 - 10.00 = 0.05 of 10.00 = 0.5%, within the 1% guardband
verdict: the ECO passes both regression checks
the added unit: 1 spare NAND2 at 2 units minus 1 removed INV at 1 unit = 2 - 1 = 1 unitNo. Function is proven, but the quality contract is missed, so the patch needs rework or signed-off acceptance.
The cone partition may be wrong. Full-design proof covers the edit and the assumption together.