D854 — A pair linked both ways is not a duplicate candidate

Decision. duplicate_candidate skips a pair when each page links the other in its body (any link outside code; k.Links(), D150/D241/D310). A satellite and its parent, and titles made of the same words (stop-words and order dropped), never close on links; for the latter the action asks to retitle one first. In a resource group the item stays while any similar pair is open.

Why. The action advised "link them saying how they differ", yet linking never closed the item: the only exit was a page-wide lint_ignore that also hid every later pairing of that page (the cost D345 removed for series pages). Mutual links mean someone edited both pages; a satellite and its parent link each other by construction; a link does not tell same-word titles apart in search or the index. It costs one more graph lookup per candidate pair.

Alternatives rejected. A one-way link closes it: the duplicate's own "see also" would silence it, often in the commit that created it. Reword the action and document lint_ignore: keeps the page-wide dismissal, and a map-level ignore would hide duplicates map-wide (D159). A pair-scoped dismissal syntax: new grammar, staleness and concept_move rewrite, the review state D298 declined.

Consequences. Pairs already linked both ways stop appearing; existing lint_ignore: [duplicate_candidate] keeps working, no migration. Removing either link, or titles converging to the same words, reopens the pair. Each rule is pinned by its own case in TestReviewDuplicateLinkedPair. Other kinds with a dismissal wider than the item (repeated_fact, map_naming, lint_judgement) are a follow-up of #834.