nx_drysweep.nx

nx_drysweep.nx -- WHAT DOES nx_gate_dry_apply ACTUALLY SAY ABOUT EVERY GATE? The D001 lane has been sized from a constant nobody measured: dry_apply's own SKIP text asserts "the fails==0 family, 36 gates" and nothing in the tree derives it (debt 1786236085). dry_apply has NO sweep verb -- its usage is `<gate> <out-candidate-path>` -- so the census could not be taken. This is that census. ★COMPOSES THE RULER, NEVER RE-IMPLEMENTS IT. It FORKS nx_gate_dry_apply per gate and only TALLIES the answer, exactly as nx_forkgrade forks nx_srcdiff per pair so that exactly ONE line-differ exists in the estate. Re-deriving dry_apply's classification here would create a second, silently-diverging judge -- and a census that disagrees with the tool it describes is worse than no census. ★DURABLE AS IT GOES. Rows APPEND to knowledge/status/drysweep.log per gate, before any summary. A long sweep that buffers its output has NO partial results: an interruption costs everything, not the remainder. Resume with <skip>. ★NO SILENT CAPS: <max> bounds the run and the DEFERRED remainder is printed. A cap reached in silence becomes a measurement nobody knows is partial. nx_drysweep <max> [skip] [dir] exit: 0 swept clean | 1 at least one ERROR/TIMEOUT (the tally is still printed) | 2 usage 3 NO-CONCLUSION (dry_apply unresolvable, or the gate dir cannot be opened) license_tier: ORIGINAL No hw writes (Rule 26).