nx_mmbench.nx

nx_mmbench.nx -- THE MEDIA-MANAGER + COMPANION CAPABILITY RULER. WHY: the ecosystem had NO product-grade ruler for either plane. nx_media_census grades the codec/room subsystem; nx_companion_census grades us against the OLD Elder AI platform. Neither answers "how far is our media manager + companion from the products people actually use". Without that number the build order is TASTE. This organ makes it ARITHMETIC. RULER = 12 REAL reference products (Jellyfin/Plex/Emby/Sonarr/qBittorrent/Kavita/yt-dlp/Immich for media; Character.AI/Replika/SillyTavern/ChatGPT-memory for companion). Each declares the capability vector it actually ships. We score OUR organs per capability and COMPUTE (never assert) coverage + a demand-ranked GAP QUEUE. That queue IS the build order. HONEST BY CONSTRUCTION (the anti-author-optimism teeth): - every HAVE must NAME an evidence artifact, and this organ STATS IT AT RUN TIME. A HAVE whose artifact is absent is LIAR-KILLED to GAP -- not warned about, DOWNGRADED. unproven = ABSENT. - a HAVE with NO named evidence is UNGROUNDED and capped at PARTIAL. You cannot assert a win. - coverage is COMPUTED from the surviving rows, so a liar-kill MOVES THE HEADLINE DOWN. - only capabilities some reference product actually ships count toward the denominator, so we cannot inflate the score by inventing axes nobody else has. - scan honesty (law L011): caps_scanned/titles_scanned/coverage_complete are emitted, never implied. Self-contained (imports only nx_syscalls) so it runs on the NAS as an MCP tool. license_tier: ORIGINAL expect_exit: 0