I argue that strict negative concord items (NCIs) are licensed by standard
(Downward) Agree. NCIs bear a [uNeg] feature which probes its c-command
domain. If no [iNeg] goal is available, an NCI moves to the closest
position which enables licensing. By virtue of this NCIs are either
base-generated or move outside of the scope of [iNeg] negation. I provide
evidence for this analysis from Russian with a special attention to the
fragment answers data. I propose that either the polarity mismatch in
ellipsis or separate deletion of negation from PF after ellipsis should be
allowed. I also show that the Agree analysis straightforwardly accounts for
ambiguous fragment answers to negative questions which are problematic for
the analyses which assume licensing in the scope of covert negation.