nx_refusal_shape_lib.nx

nx_refusal_shape_lib.nx -- DOES THIS REFUSAL NAME ITS SUBJECT AND ITS REMEDY? FM11 of /compare/failmodes: the MECHANIZABLE half of the tenth cell, LOUD-AND-CORRECT. The taxonomy lane established that EMITTER QUALITY and CALLER RESPONSE are two INDEPENDENT AXES and proved it with one instance that scores on both at once: nx_fs read declares its truncation cap AND its remedy in its last line (emitter CORRECT) while the reader filtered that line out and filed it as a critical defect (caller IGNORED). A single axis keyed on "did the caller obey" would file that instance under LOUD-AND-IGNORED and exclude the very case that motivated the cell. THIS LIBRARY OWNS THE EMITTER HALF ONLY, AND THAT SCOPE IS DELIBERATE. The caller half -- did anyone act on the refusal -- is CENSUS-ONLY over the call record and is explicitly OUT OF SCOPE here, because a refusal that callers routinely retry through is LOUD-AND-IGNORED however well written it is. No function in this file may be read as evidence about a caller. WHY A LIBRARY AND NOT A SECOND ORGAN. The census verb lives in nx_failclass (fc_refusal_shape) and the gate lives in nx_refusal_shape_gate, and the two MUST agree about what "shaped" means. They agree BY CONSTRUCTION rather than by discipline: there is exactly ONE classifier (rs_bucket), ONE block builder and ONE vocabulary loader, and both consumers import them. THE UNIT IS THE COMPOSED EMISSION, NOT THE STRING LITERAL, AND THAT IS THE WHOLE DESIGN. Measured on the canonical guard before a line was written: nx_tools_api.nx builds the truncation envelope from THREE pieces -- b = ta_cat(jb, b, " [NX-TRUNCATED OUTPUT-IS-PARTIAL capture_cap=" as *u8) b = ta_catn(jb, b, xcap) b = ta_cat(jb, b, " -- narrow the query or re-run with _async=1]" as *u8) -- so the refusal marker and its remedy live in DIFFERENT LITERALS with a non-literal line between them. A per-literal checker would flag a KNOWN-GOOD GUARD, and * A SHAPE CHECKER THAT FLAGS A KNOWN-GOOD GUARD IS WORSE THAN NONE, BECAUSE SOMEONE WILL ACT ON IT. * So a MESSAGE here is a BLOCK: a run of consecutive source lines carrying string literals, tolerating up to max_gap_lines literal-free lines, broken HARD by a blank line, a comment line, or a line whose first non-space character closes a scope. The brace break is what stops a block borrowing an unrelated neighbour's remedy, and it is bite-proven in the gate. COMMENTS ARE NOT SCANNED. A source-scanning detector finds the PROSE DESCRIBING the defect and accuses the author who documented it -- prose is source bytes too. Comment lines are hard breaks and no literal is ever harvested from one.