nx_discovery_selfheal_beat.nx

nx_discovery_selfheal_beat.nx -- makes the discovery-selfheal check ACTUALLY self-heal. THE DEFECT THIS EXISTS TO KILL (measured 2026-07-30): the sweep row ran `nx_toolreg_reconcile heal`, a verb named "heal" that only DIAGNOSES -- it prints `no-schema-residue=N needs-author-schema: ...` and exits 1. Closing it needs a THREE-COMMAND MANUAL SEQUENCE (schema_backfill -> reconcile -> heal), so every time any lane shipped a tool without a schema header the row went RED and STAYED red until a human noticed. It sat red for 24+ consecutive cycles at 0 permille. Clearing the backlog by hand does NOT fix it: within one hour of clearing 14, five MORE arrived (nx_meshprofile, nx_routeguard, nx_wirecensus, nx_arousal_skin_gate, nx_arousal_mesh_gate) from other lanes shipping normally. A treadmill is not a defect in the lanes -- it is a missing loop. So this beat RUNS the loop the check was only ever reporting on: 1. ./nx_schema_backfill.elf harvest schemas from source headers (idempotent) 2. ./nx_toolreg_reconcile.elf publish them into the registry 3. ./nx_toolreg_reconcile.elf heal judge, and EXIT WITH ITS CODE Step 3's rc is the beat's rc, so the rail still judges the real invariant -- we heal first, then report honestly. If a residue SURVIVES the backfill (an organ whose source has no header at all) the beat still goes RED and NAMES it, because that one genuinely needs a human to author it. Rule 10 idempotent: backfill is idempotent by contract; reconcile skips already-registered rows. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0