nx_swapguard_beat.nx

nx_swapguard_beat.nx -- SWAP-EXHAUSTION TRIPWIRE (2026-07-29, seq1256 lineage). Today's mgmt_api and tools_api wedges were both preceded by HOURS of exhausted swap (12-196kB free) with no alarm anywhere -- long-running seg-store consumers leak into swap (seq905/947 class) and then park in uninterruptible sleep. This beat reads /proc/meminfo SwapFree every scheduler cycle and, on breach, (a) appends an evidence line and (b) files a sev8 debt row THROUGH nx_debt whose desc is CONSTANT -- nx_debt add is content-idempotent (banked law 07-20), so hourly re-alarms dedupe to ONE standing row and the board (which gates nx_ws_cycle work-emission) becomes the alarm surface. ALARM-ONLY by design: auto-recycle needs the sanctioned restart paths (seq1256) and graduates only after this tripwire proves calibrated. scheduler contract: jobs.tsv fork+execs with NO args. argv[1]=floor_kb override (testing). floor default 2097152 kB = 2GiB, DERIVED: swap total ~24GiB; every observed wedge sat at <200kB for hours, fresh-boot free is >20GiB -- 2GiB (~8%) alarms early while never false-firing on a healthy host (rule-11: promote to conf when field-tuned). trend line: <epoch>\tSwapFree=<kb>\t<OK|ALARM>\n -> knowledge/status/swapguard_trend.jsonl license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0 healthy, 1 alarm