[dmarx]

Why

In nucleation, the QUESTION scheme has active: Open. When an Open parent question is split, its answered part becomes a child question with status Answered (nucleation's ADR-049). The parent then cites that child, and the retired-citations check warns on every such citation, because any status other than active counts as retired. The owner's rule is that an Open question citing an Answered one is legitimate. A citation from anywhere else should still warn: it either still treats the question as open, or it should cite the CLAIM or THEORY that answers it. Today the only remedy is an inactive-ok comment at each site.

What

lint.inactive_ok is the config-level form of the inactive-ok directive. It is a list of rules written in the existing pattern syntax (patterns.lark), with a reserved hop type cites:

lint:
  inactive_ok:
    - match: "(:QUESTION {status: 'Open'})-[:cites]->(:QUESTION {status: 'Answered'})"
      reason: a split parent cites the Answered child split off it

A citation of a retired document is sanctioned, and needs no acknowledgement, when the document it appears in matches the source node and the cited document matches the target node. Properties match any front-matter field (status, role, warrant and so on), as they do in patterns.X. The match: value has to be quoted in YAML, because an unquoted { starts a mapping.

Refused at load, each naming lint.inactive_ok[N]:

• a value that is not a list, an entry that is not a mapping, or keys other than match and reason;
• a missing or blank reason, mandatory for the same reason directive reasons are;
• zero hops or more than one;
• a hop typed anything but cites alone;
• variable length, or WHERE NOT;
• an unlabelled node;
• an undeclared scheme, a property that scheme can't hold, or MATCH and similar keywords.

Labels and properties are checked by the same code that checks patterns.X. The grammar is not forked: patterns.parse and check_patterns each take an optional argument naming where the pattern was declared.

Dead rules (DP-015). A rule that sanctions nothing is reported as a stale-directives row, with its pattern and reason. An inactive-ok comment that a rule has made redundant is reported stale, with a message naming the rule.

Internals. A matching site emits sanctioned(I), and retired(I) in acknowledgements.lp now requires not sanctioned(I). Citation gains sanctioned and a loud property, which flagged(), ack.py and adr_pending.py use to decide what is unacknowledged. Sanctioned sites are not counted as acknowledged, because they aren't suppressions. The export's excused column keeps its meaning, "a directive acknowledges it".

History of this PR

The first commit added a per-scheme citable_from: {Answered: [QUESTION:Open]}. The owner asked for something more general, since status words are specific to one scheme and that form could only express status. The second commit replaces it with the pattern rules above. The ADR records citable_from as a rejected alternative, alongside:

• active as a list, which would also silence a document still citing a now-Answered question as open;
• per-site annotations only, which is the status quo;
• a global "counts as in force" flag per status.

Record

• ADR ADR-tmpfu1s6, numbered at merge.
• A changelog fragment.
• A devlog entry recording the guard being fired on a real case (DP-006). On a scratch copy of nucleation with the rule above:
  • the 13 split parents' inactive-ok comments went stale, each naming lint.inactive_ok[0];
  • unacknowledged rows stayed at 480;
  • 34 sites moved from "acknowledged" to "sanctioned".
• config_doc gains a section, so the configuration reference documents the key.

Tests

• tests/test_inactive_ok_rules.py, 27 tests; the 26 written first failed before the change.
• Full suite: 1881 passed, 33 skipped.
• luria lint gives the same finding classes and counts as main.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KC8NN4NtUf8sd5unC6dpFV