nx_execsurface.nx

nx_execsurface.nx -- WHICH SURFACE RUNS THIS PROGRAM? The execution-surface census for PROMOTED organs (2026-08-23, compare CE2, operator: no false ceilings). THE FALSE CEILING THIS CLOSES: the adoption ladder's INVOKED axis reads knowledge/status/actlog.jrnl, which records MCP-surface invocations ONLY. A program the clock plane fires nightly, a cron row, a daemon supervisor, the gate roster, or a ./x.elf fork from another organ therefore read REGISTERED-DARK ("no MCP invocation on record") while executing every day -- measured on the live graphics board: 12 of 19 PARTIAL rows, among them organs gates fork on every beat. nx_gateadjudicate solved exactly this for GATES by joining the gate population to the surfaces that CAUSE EXECUTION; nobody did it for programs. THE MCP REGISTRY IS NOT THE ONLY DOOR, AND A LADDER THAT ONLY WATCHES ONE DOOR CALLS EVERY OTHER ENTRANT DARK. THE EDGE DEFINITION IS THE QUESTION (gateadjudicate's law, measured 08-07: four definitions gave 75 / 0 / 2,043 / 30). A program counts as ON A SURFACE only through something that CAUSES execution: CLOCK the live clocksched- plane names its .elf in the organ column, split at spaces so an organ named in another row's ARGUMENTS (nx_sov_build_run.elf <gate>) counts too CRON a cron.reg row names <n>.elf, <n>.sh or <n>.cron.sh (shell wrappers spawn the organ) PLAN a clock row runs nx_plan_run.elf <id>: every tool the rows of knowledge/store/plan-<id>- name is credited (2026-09-17; the clock row is the cause, the plan the route -- workflows-as-data are a surface, and before this every plan-run organ read REGISTERED-DARK, measured on nx_search_ship) DAEMON a daemons.reg row's argv names <n>.elf (the supervisor keeps it alive) ROSTER a row of a ROSTER CONF names it (a beat of nx_gate_roster_run forks it): the runner's default knowledge/gateroster.conf and every conf a live clock row runs through nx_gate_roster_run beat, whatever journal that beat writes, derived from the plane by nx_rosterconf_lib and read with the beat's own row grammar (2026-09-18: the heavy roster's gates read roster=0 while its beat ran them daily; a later roster is read by declaring its clock row, with no edit here). A slow beat writing its own journal is credited like any other: the close ruler reads each roster's evidence from that roster's journal; the stamp still counts those beats (roster_beats_other_journal) ACTLOG the MCP tail window mentions it (the one door the ladder already watched) FORK another organ's SOURCE carries the string literal "./<n>.elf", "../<n>.elf" or "_offc/<n>.elf" AND that caller is itself on a surface -- 2-hop, iterated to a fixed point, so a beat-run organ that forks a helper that forks a third credits the third. A literal in a caller that is NOT on any surface is a REF: something wants it, nothing runs it. A MERE MENTION IS NOT INVOCATION. The literal must open a string ("./x.elf), so prose in a comment cannot credit a fork. ROWS name<TAB>clock<TAB>cron<TAB>daemon<TAB>fork<TAB>roster<TAB>actlog<TAB>first_surface clock/cron/daemon/roster/actlog are 0|1; fork is "-" or the first crediting caller's basename; first_surface is clock|cron|daemon|roster|actlog|plan|fork:<caller>|ref:<caller>|- (the token a board prints) LAST LINE (the canonical stamp, anchored BY POSITION; consumers never grep the rows for it): # asof=<epoch> organs=N on_surface=A ref_only=B dark=C sources=clock:<rows>,cron:<rows>,daemon:<rows>, roster:<rows>,roster_confs:<read>,roster_beats:<n>,roster_beats_other_journal:<n>,actlog:<bytes>,fork_files:<n>,