Mark each attachment point with a numbered dummy atom, e.g. [*:1], [*:2].
Not pinned to one attachment point — the engine tries every open ring position on the scaffold and dedupes chemically identical results. Applies whether your indexed R-groups above came from manual entry or patent-text extraction.
Leave blank to use your plan's full limit.
When a request is truncated, pick a chemically diverse subset (Morgan fingerprints + MaxMin) instead of just the first ones found.
Off by default: every result is annotated (PAINS, reactive groups, high LogP) but nothing is removed, so patent-scope coverage stays complete. Turn on to drop flagged structures instead.
e.g. "If R2 is not H, R5 cannot be N." Positions 1..N refer to the indexed R-group positions in step 2; if you also have roaming R-group roles, they're numbered right after (e.g. 2 indexed positions + roaming role 1 = position 3). Roaming positions match on which fragment was picked for that role, not on which scaffold atom it landed on.