search-translation
A deny rule: it refuses a text search whose pattern a graph query provably answers with the same entities, and names what to run instead.
What it catches
A text search whose pattern a graph query provably answers with the same entities.
Why
The pattern is compiled in the tool's own dialect (BRE, ERE or fixed) and run against the graph's ids when the command is judged, so the deny names a query that was checked rather than one that looks equivalent. Three shapes qualify. A pattern that can only match MGS codes (MGS30[23], MGS30.., MGS302[0-9]\|MGS303[0-9]), over any path in the workspace, becomes magus query kind=diagnostic 'id=~^diagnostic:...$', and a single literal code keeps symbol-search's magus explain diagnostic:<code>. A pattern selecting every Markdown heading of the files searched (^#, ^#\+), when those lines match the section nodes the graph holds file for file and none sits in a code fence, becomes magus query kind=docsection 'id=~^docsection:<file>#'. A search of a magusfile whose every hit declares a target the graph holds becomes magus explain target:<project>:<name>. Anything else stays silent: -i, -v, -c, -l, -x, context flags, a level-specific heading pattern, a BZZ code, a line anchor on a code, a heading inside a fence, one hit that is a call or a comment, stdin, or a tree outside the workspace. Measured 2026-09-24: 14,773 searches, 45% of them alternations, and graph verbs used about 50 times less than grep.
Seeing it
A verdict names its rule in brackets, which is how you got here:
deny [search-translation]: ...
magus describe rule search-translation prints the same entry at a terminal, and
magus describe rules lists every rule this workspace enforces.