{
 "schema": "cain.e89.e88_reproduction.v1",
 "e88_commit": "f2d12e8b6",
 "method": "recompute from raw rows in the E88 worktree; summaries are not trusted",
 "checks": [
  {
   "id": "ADV-P1-ARMS",
   "claim": "phase-1 correct per arm (out of 15)",
   "reported": {
    "A": 3,
    "B": 4,
    "C": 4,
    "D": 5,
    "E": 4,
    "F": 7,
    "G": 4,
    "H": 5
   },
   "recomputed": {
    "A": 3,
    "B": 4,
    "C": 4,
    "D": 5,
    "E": 4,
    "F": 7,
    "G": 4,
    "H": 5
   },
   "source": "artifacts/e88/advantage/rows_phase1.jsonl",
   "verdict": "REPRODUCED",
   "note": ""
  },
  {
   "id": "ADV-P2-ARMS",
   "claim": "phase-2 correct per arm (out of 15)",
   "reported": {
    "C": 4,
    "H": 0
   },
   "recomputed": {
    "C": 4,
    "H": 0
   },
   "source": "artifacts/e88/advantage/rows_phase2.jsonl",
   "verdict": "REPRODUCED",
   "note": ""
  },
  {
   "id": "ADV-DENOM",
   "claim": "'30 held-out tasks' = 15 distinct phase-1 + 15 distinct phase-2 tasks",
   "reported": {
    "phase1": 15,
    "phase2": 15,
    "overlap": 0,
    "distinct": 30
   },
   "recomputed": {
    "phase1": 15,
    "phase2": 15,
    "overlap": 0,
    "distinct": 30
   },
   "source": "rows_phase1.jsonl, rows_phase2.jsonl",
   "verdict": "REPRODUCED",
   "note": "The '30 held-out tasks' headline is a suite size; every per-arm table uses n=15 per phase and no arm except C and H was ever run on all 30."
  },
  {
   "id": "ADV-P1-C-vs-F",
   "claim": "planning arm F vs C: +20 pp, p=0.375",
   "reported": {
    "diff_pp": 20.0,
    "p": 0.375
   },
   "recomputed": {
    "diff_pp": 20.0,
    "p": 0.375
   },
   "source": "rows_phase1.jsonl",
   "verdict": "REPRODUCED",
   "note": ""
  },
  {
   "id": "ADV-P1-C-vs-H",
   "claim": "full stack H vs C: +6.7 pp, p=1.0",
   "reported": {
    "diff_pp": 6.67,
    "p": 1.0
   },
   "recomputed": {
    "diff_pp": 6.67,
    "p": 1.0
   },
   "source": "rows_phase1.jsonl",
   "verdict": "REPRODUCED",
   "note": ""
  },
  {
   "id": "ADV-P1-F-vs-H",
   "claim": "full stack did not outperform planning-only (F 7 vs H 5)",
   "reported": {
    "F": 7,
    "H": 5
   },
   "recomputed": {
    "F": 7,
    "H": 5
   },
   "source": "rows_phase1.jsonl",
   "verdict": "REPRODUCED",
   "note": "Direction reproduces; paired F vs H {'b_control_only': 3, 'c_treatment_only': 1, 'p': 0.625} is not significant at n=15."
  },
  {
   "id": "ADV-P2-C-vs-H",
   "claim": "phase 2: C 4/15, H 0/15, p=0.125",
   "reported": {
    "diff_pp": -26.67,
    "p": 0.125
   },
   "recomputed": {
    "diff_pp": -26.67,
    "p": 0.125
   },
   "source": "rows_phase2.jsonl",
   "verdict": "REPRODUCED",
   "note": ""
  },
  {
   "id": "ADV-P2-DIAG",
   "claim": "phase-2 memory regression diagnosis",
   "reported": "procedure line + verification retry consumed budget",
   "recomputed": {
    "H_verification_failed_tasks": 9,
    "H_skill_used_tasks": 13,
    "H_output_tokens": 858,
    "C_output_tokens": 337,
    "H_tool_ok": 6,
    "C_tool_ok": 3,
    "H_families_lost": [
     "date_diff",
     "ratio_scale",
     "weighted_mean"
    ]
   },
   "source": "rows_phase2.jsonl",
   "verdict": "DIAGNOSED",
   "note": "Recomputed from raw rows; see audit for interpretation."
  },
  {
   "id": "ADV-SEEDED",
   "claim": "memory arm E seeded from disjoint experience",
   "reported": "disjoint",
   "recomputed": {
    "arithmetic_chain": "math",
    "base64_decode": "security",
    "checksum": "infrastructure",
    "compound_interest": "business",
    "date_diff": "planning",
    "hash_prefix": "security",
    "interpolate": "science",
    "percent_change": "business",
    "sort_median": "planning",
    "string_normalize": "software",
    "subnet_hosts": "infrastructure",
    "unit_conversion": "engineering",
    "weighted_mean": "research"
   },
   "source": "seeded_experience.json",
   "verdict": "INSPECTED",
   "note": "Experience seeds are task families, so the 'memory' arm reuses a procedure for the same family on a new instance: in-distribution reuse, not cross-family learning."
  },
  {
   "id": "MODEL-SCALE",
   "claim": "full stack -10 pp at 0.5B and 1.5B, 0 pp at 135M",
   "reported": {
    "HuggingFaceTB/SmolLM2-135M-Instruct": {
     "C": 0,
     "H": 0,
     "n": 10,
     "diff_pp": 0.0
    },
    "Qwen/Qwen2.5-0.5B-Instruct": {
     "C": 1,
     "H": 0,
     "n": 10,
     "diff_pp": -10.0
    },
    "Qwen/Qwen2.5-1.5B-Instruct": {
     "C": 6,
     "H": 5,
     "n": 10,
     "diff_pp": -10.0
    }
   },
   "recomputed": {
    "HuggingFaceTB/SmolLM2-135M-Instruct": {
     "C": 0,
     "H": 0,
     "n": 10,
     "diff_pp": 0.0
    },
    "Qwen/Qwen2.5-0.5B-Instruct": {
     "C": 1,
     "H": 0,
     "n": 10,
     "diff_pp": -10.0
    },
    "Qwen/Qwen2.5-1.5B-Instruct": {
     "C": 6,
     "H": 5,
     "n": 10,
     "diff_pp": -10.0
    }
   },
   "source": "artifacts/e88/models/models.json M2_paired.rows",
   "verdict": "REPRODUCED",
   "note": "-10 pp is one task out of 10; it is direction, not a measured disadvantage."
  },
  {
   "id": "TEAMS",
   "claim": "team strategies correct / false accept / false reject (n=10)",
   "reported": {
    "single": [
     2,
     8,
     0
    ],
    "identical_three": [
     2,
     8,
     0
    ],
    "diverse_three": [
     2,
     8,
     0
    ],
    "specialist_verifier": [
     2,
     8,
     0
    ],
    "heterogeneous_verifier": [
     1,
     0,
     4
    ],
    "team_memory": [
     1,
     9,
     0
    ]
   },
   "recomputed": {
    "diverse_three": [
     2,
     8,
     0
    ],
    "heterogeneous_verifier": [
     1,
     0,
     4
    ],
    "identical_three": [
     2,
     8,
     0
    ],
    "single": [
     2,
     8,
     0
    ],
    "specialist_verifier": [
     2,
     8,
     0
    ],
    "team_memory": [
     1,
     9,
     0
    ]
   },
   "source": "artifacts/e88/teams/teams.json rows",
   "verdict": "REPRODUCED",
   "note": ""
  },
  {
   "id": "FENCE",
   "claim": "0 effects after the fence at 10/50/100/250 workers",
   "reported": {
    "10": {
     "effects": 3510,
     "after_fence": 0,
     "pids": 10,
     "denied_lines": 3563
    },
    "50": {
     "effects": 3120,
     "after_fence": 0,
     "pids": 50,
     "denied_lines": 3100
    },
    "100": {
     "effects": 2643,
     "after_fence": 0,
     "pids": 100,
     "denied_lines": 2978
    },
    "250": {
     "effects": 2434,
     "after_fence": 0,
     "pids": 250,
     "denied_lines": 2922
    }
   },
   "recomputed": {
    "10": {
     "effects": 3510,
     "after_fence": 0,
     "pids": 10,
     "denied_lines": 3563
    },
    "50": {
     "effects": 3120,
     "after_fence": 0,
     "pids": 50,
     "denied_lines": 3100
    },
    "100": {
     "effects": 2643,
     "after_fence": 0,
     "pids": 100,
     "denied_lines": 2978
    },
    "250": {
     "effects": 2434,
     "after_fence": 0,
     "pids": 250,
     "denied_lines": 2922
    }
   },
   "source": "artifacts/e88/scaling/fence_*/effects_*.jsonl",
   "verdict": "REPRODUCED",
   "note": "Recomputed by comparing every raw effect timestamp with the recorded effective instant. 'denied_lines' sums the per-worker denial counters (one integer per file)."
  },
  {
   "id": "ENDURANCE-1H",
   "claim": "one-hour run: 71,697 tasks, 0 duplicates, ~3,603 s",
   "reported": {
    "tasks": 71697,
    "duplicates": 0,
    "span_s_min": 3590
   },
   "recomputed": {
    "tasks": 71697,
    "duplicates": 0,
    "span_s_min": 3590
   },
   "source": "artifacts/e88/scaling/endurance_3600/effects_*.jsonl",
   "verdict": "REPRODUCED",
   "note": "raw first->last effect span 3599.9 s; the JSON keeps only the last [{'alive': 4, 't': 2967.0, 'tasks': 59087, 'verified': 59087}, {'alive': 4, 't': 3028.1, 'tasks': 60306, 'verified': 60306}, {'alive': 4, 't': 3089.2, 'tasks': 61525, 'verified': 61525}, {'alive': 4, 't': 3150.3, 'tasks': 62743, 'verified': 62743}, {'alive': 4, 't': 3211.6, 'tasks': 63962, 'verified': 63962}, {'alive': 4, 't': 3273.1, 'tasks': 65168, 'verified': 65168}, {'alive': 4, 't': 3334.3, 'tasks': 66390, 'verified': 66390}, {'alive': 4, 't': 3395.7, 'tasks': 67609, 'verified': 67609}, {'alive': 4, 't': 3457.0, 'tasks': 68830, 'verified': 68830}, {'alive': 4, 't': 3518.4, 'tasks': 70050, 'verified': 70050}, {'alive': 4, 't': 3579.7, 'tasks': 71271, 'verified': 71271}, {'alive': 0, 't': 3601.8, 'tasks': 71697, 'verified': 71697}] checkpoints, not all 60; PSS was not captured (0.0 is a gap)."
  }
 ],
 "counts": {
  "REPRODUCED": 11,
  "DIAGNOSED": 1,
  "INSPECTED": 1
 }
}
