nx_contentdiff.nx

nx_contentdiff.nx -- THE PROMOTE RULER. Answers one question: does the candidate binary still contain everything the LIVE binary contains? WHY (debts 1785452162, 1785526315, 1785526809): the ecosystem has ~703 deployed organs awaiting a rebuild, and the gate guarding that queue was BYTE SIZE. Size is uninformative IN BOTH DIRECTIONS, proven twice by measurement: SHRINK is not regression -- 16 live organs had LARGER .prev files, yet the smaller live binaries were string SUPERSETS (lost_from_live=0). Restoring the bigger ones would have BEEN the regression. GROWTH is not improvement -- nx_law_warden rebuilt GREW 161951 -> 186696 bytes while losing three whole detectors (laws_measured 11 -> 8, enforced_permil 1000 -> 727). A gate keyed on bytes is therefore wrong about half the time and cannot tell you which half. THE RULER: printable content. Every printable run of >= CD_MINLEN bytes in the LIVE binary must still be findable somewhere in the candidate. Anything missing is capability the rebuild would DESTROY. Searching the candidate's RAW BYTES (not a re-extracted string set) is deliberate and stricter: a string that survived but got tokenized differently still counts as present, so a MISS is a real miss. FAIL-CLOSED: unreadable input is RED, never "clean". Verdict GREEN iff lost_from_live == 0. NOT DEDUPED, and it says so in the output: runs are counted as encountered, so the counts are occurrence counts and lost>0 is the signal -- declaring that beats a silent, prettier number. nx_contentdiff <live-elf> <candidate-elf> license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0