nx_zeroprint.nx

nx_zeroprint.nx -- THE ZERO-PRINTS-AS-NUL DETECTOR AND REPAIRER. FOUND 2026-08-26 while adding a DNS diagnostic: a field printed "ancount= " with nothing after it. The value was 0. The shape, repeated by hand across the estate, is a decimal printer that writes its digits into one buffer and reverses them into another: if m == 0 { b[0] = 48 as u8; k = 1 } <-- '0' written to b let t: *u8 = sys_mmap(24) while m > 0 { t[k] = ...; k = k + 1 } <-- digits written to t while i < k { b[i] = t[k-1-i]; i = i+1 } <-- reversal reads t, OVERWRITING b[0] With m == 0 the digit loop never runs, so t is untouched mmap memory (zero-filled) and the reversal copies a NUL byte over the '0'. A zero therefore prints as an INVISIBLE BYTE. That is the worst possible failure direction for a diagnostic: the field reads as ABSENT rather than as the value zero, and "absent" and "zero" demand opposite conclusions from whoever is reading it. WHY THIS IS AN ORGAN AND NOT A ONE-OFF SWEEP: the first census of this was written in Python -- shell-shaped work that counted, classified and judged, which is precisely the shape the estate forbids, and it published a wrong number (28) because it could not see the safe else-form. The count belongs in a NishiLang organ that can be re-run, gated, and pointed at either tree. THE CLASSIFICATION IS A PARTITION AND IT IS PRINTED, because three of the four buckets look identical to a careless matcher: EARLY-RETURN if m == 0 { b[0] = 48; sys_write(...); return 0 } -- prints and leaves. SAFE. ELSE-FORM if m == 0 { ... } else { ...digit loop + reversal... } -- reversal is inside the else and cannot run when m == 0. SAFE. This is the bucket the Python missed, and missing it inflated the defect count by six. SAME-BUFFER the zero branch writes the buffer the digit loop writes. The reversal reads back a real '0'. SAFE -- this is the majority and the shared printer in nx_syscalls.nx. DEFECTIVE zero branch writes X, digit loop writes Y, X != Y. The NUL case above. REPAIR is a pure statement reorder, not a rewrite: the declaration of Y already sits between the zero branch and the digit loop, so moving the zero branch BELOW that declaration and retargeting it at Y makes the reversal read a real '0'. Nothing else about the function changes. NO SEARCH WINDOW CONSTANT. A "look ahead N bytes" bound would be a guess and would silently under-report on a long function. The digit loop is searched for only up to the START OF THE NEXT FUNCTION, which is exact and needs no number. verbs: scan <root> census only, exit 1 if any DEFECTIVE remain (gate-friendly)