{
  "schema_version": 1,
  "generated_at": "2026-07-13",
  "decision": "Advance perception, task-local scoring, and exact evidence retrieval. Do not promote the current deterministic Story State.",
  "scope": {
    "dataset": "sochastic/Re-Verse",
    "dataset_revision": "aadc32087f3aa9b5eca0bfd40466d3664a8d4a84",
    "pages": 308,
    "chapters": 11,
    "npp_rows": 433,
    "test_targets_context_5": 38,
    "test_targets_context_10": 28,
    "test_chapters": [10, 11],
    "single_title": true,
    "test_was_inspected_during_protocol_repair": true
  },
  "inference": {
    "label": "conditional_descriptive",
    "method": "within-chapter circular moving-block percentile bootstrap",
    "block_size": 4,
    "draws": 10000,
    "warning": "Intervals condition on two observed chapters of one inspected title. They are not title-level, population-generalization, or confirmatory confidence intervals."
  },
  "npp": {
    "candidate_count": 10,
    "chance": {"top1": 0.1, "mrr": 0.2928968253968254},
    "frozen_siglip_similarity": {
      "input_status": "vision_only",
      "context_5": {"top1": 0.2894736842105263, "mrr": 0.4632205513784462},
      "context_10": {"top1": 0.21428571428571427, "mrr": 0.3869897959183673}
    },
    "siglip_plus_local_oracle_text_similarity": {
      "input_status": "privileged_incremental_utility_control",
      "context_5": {"top1": 0.39473684210526316, "mrr": 0.5630847953216375},
      "context_10": {"top1": 0.39285714285714285, "mrr": 0.5591695011337868}
    },
    "siglip_plus_matched_story_state_similarity": {
      "input_status": "privileged_incremental_utility_diagnostic",
      "context_5": {"top1": 0.21052631578947367, "mrr": 0.43434628237259815},
      "context_10": {"top1": 0.21428571428571427, "mrr": 0.4237528344671202}
    },
    "local_text_minus_matched_state": {
      "candidate_and_siglip_prefix_features_identical": true,
      "context_5": {
        "top1_difference": 0.18421052631578946,
        "top1_interval95": [0.05263157894736842, 0.3157894736842105],
        "mrr_difference": 0.1287385129490393,
        "mrr_interval95": [0.037290361319966596, 0.22526133040935678]
      },
      "context_10": {
        "top1_difference": 0.17857142857142858,
        "top1_interval95": [0.03571428571428571, 0.2857142857142857],
        "mrr_difference": 0.13541666666666666,
        "mrr_interval95": [0.03283588435374149, 0.23018990929705213]
      }
    },
    "qwen2_5_vl_7b_balanced": {
      "model_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
      "protocol": "Each task places gold exactly once at every A-J position across ten seeds; negatives are independently shuffled.",
      "validation": {
        "context_5": {"top1": 0.13404255319148936, "mrr": 0.32913880445795335, "parse_rate": 0.9914893617021275},
        "context_10": {"top1": 0.1, "mrr": 0.3158526383526384, "parse_rate": 1.0}
      },
      "test": {
        "context_5": {"top1": 0.17105263157894735, "top1_interval95": [0.11315789473684215, 0.22894736842105257], "mrr": 0.3714254385964912, "parse_rate": 0.9973684210526316},
        "context_10": {"top1": 0.17142857142857143, "top1_interval95": [0.1142857142857143, 0.2285714285714286], "mrr": 0.36034863945578227, "parse_rate": 1.0}
      },
      "qwen_minus_siglip_top1": {
        "context_5": {"difference": -0.11842105263157894, "interval95": [-0.2789473684210527, 0.028947368421052628]},
        "context_10": {"difference": -0.04285714285714287, "interval95": [-0.17142857142857135, 0.08571428571428572]}
      },
      "mean_unique_canonical_top_candidates_across_permutations": 4.196969696969697
    }
  },
  "perception": {
    "magiv2_standard_iou_0_5": {
      "scope": "class-agnostic diagnostic; no confidence scores or AP",
      "characters": {"precision": 0.7726663689146941, "recall": 0.947945205479452, "f1": 0.8513779527559056},
      "text": {"precision": 0.08066239316239317, "recall": 0.09667093469910372, "f1": 0.08794408852649972}
    },
    "magiv2_ios_0_5_containment_only": {
      "warning": "IoS is a containment diagnostic, not IoU, AP, or a substitute detection score.",
      "characters": {"precision": 0.7887449754354623, "recall": 0.9676712328767123, "f1": 0.8690944881889763},
      "text": {"precision": 0.8317307692307693, "recall": 0.9967989756722151, "f1": 0.9068142108328481}
    },
    "ocr_proxy": {
      "pages": 4,
      "teacher_turns": 11,
      "ocr_strings": 25,
      "mean_best_token_f1_per_teacher_turn": 0.8327637200898579,
      "teacher_turn_coverage_at_token_f1_0_5": 1.0,
      "mean_best_token_f1_per_ocr_string": 0.47271549028510096,
      "warning": "Adapted-prose lexical proxy; not OCR ground truth, CER, WER, precision, or recall."
    }
  },
  "reproducibility": {
    "tests_passed": 51,
    "ruff": "passed",
    "model_revisions": {
      "siglip": "9fdffc58afc957d1a03a25b10dba0329ab15c2a3",
      "qwen2_5_vl_7b": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
      "magiv2": "fbc890fec52977142e8ee00bfe26e9458b65517c"
    },
    "sha256": {
      "npp_tasks": "0c06f1b91c1a4a5e86004ab122ca028eae9ccc7b580f7a864cc9085b149f603c",
      "reverse_manifest": "b1a85a2a50c57297cc635eb0041aedc1143677e14f54a9eea63ba5c4b073479e",
      "siglip_embeddings": "ce841b7a01690afee77168bd1d00b272af577e71f5dfd93411b9986a6996b3bf",
      "npp_probe_results_v4": "4e4743754f0e3bfbb02b227fe22f02733dd36d3e87c0feaabdd6ee354924cb0c",
      "npp_probe_statistics_v4": "1bd23d47171b1e254f0da399f9f01b34ce0687befe87d6096ad54030b8dfa480",
      "qwen_balanced_validation": "d2b6ba31103c69e7aa153125558f6058bc090025e4a547db1f4bc65c4964e550",
      "qwen_balanced_validation_statistics": "b1547e0e2af0bb567fd1bb60b0944a372637288c32fef7cbb2f7e0a24a5abf07",
      "qwen_balanced_test": "50e1830ff3f510f9c44e4e230de20f57fa3f623c3baf76255baba8642ab30683",
      "qwen_balanced_test_statistics": "09d3431a4d1a2f090c1a7db0a2c257ef186a295480d001fee939197460b95993",
      "magiv2_detections": "c9ceb4f0417a09facaf8d55508d31a58320b1a38a10c4502e2e87d11458637ad",
      "magiv2_iou_metrics": "1eba80d1d9684a90146052d38cee8dc65fcf853a26689d0a86f933be01cb93e2",
      "magiv2_ios_metrics": "84eddb351bf13bd6cabe7cadb7eaf2d0403a787195d0f59148f9a5e4110639e8",
      "ocr_proxy": "2118ccc5c59d3e2eebeec665ec8ba5458d05a9419a92da2df24090d3858509c2"
    }
  },
  "rights": "No dataset images, prose, recoverable text, or model checkpoints are included in this public summary."
}