[
  {
    "id": 1,
    "topic": "Evidence and inference",
    "title": "Why Do the Theist Sides Score Lower?",
    "short": "Where the score gap comes from",
    "subtitle": "A smaller gap in a larger archive—and the limits of explaining it.",
    "stats": [
      [
        "5.74",
        "average skeptical-side advantage\nin score points / 100"
      ],
      [
        "234",
        "classified comparisons\nfrom 308 assessments"
      ],
      [
        "47",
        "new eligible comparisons\nsince September"
      ]
    ],
    "finding": "The score gap persists, but newer additions are closer.",
    "statistic": "5.74 points",
    "statisticLabel": "Average non-theist advantage · 234 comparisons",
    "explanation": "The September gap was 6.34 points; the 47 new comparisons average 3.34, bringing the combined gap down to 5.74. Skeptical sides lead in 193 assessments, religious sides in 33, with 8 ties. Support remains the largest raw scoring-area difference. Support, logic and replies contribute 74.6% of the overall gap.",
    "detail": "The new cases include meaningful counterexamples, including Meyer–Atkins on design. A creator argument is not automatically a defense of a personal religious God; those narrower burdens and explicit inclusion decisions remain visible. The narrower truth-claim set retains a 5.77-point gap.",
    "limitation": "This curated archive measures assessed performances, not worldview truth. The faith-to-debate hypothesis remains a proposed explanation, not an established psychological cause. Repeated speakers and judging preferences are not eliminated by resampling.",
    "pdf": "why-do-the-theist-sides-score-lower",
    "figure": "p1-dimensions",
    "links": [
      {
        "id": "jones-schmid-contingency-god-2022",
        "label": "Jones–Schmid: from a foundation to God"
      },
      {
        "id": "meyer-atkins-intelligent-design-science-2010",
        "label": "Meyer–Atkins: a counterexample to the average"
      }
    ],
    "method": "Reconstruct scores with official move importance, section weights and rounding. Subtract the religious-side score from the skeptical-side score within each debate, then average with one vote per debate. Draw whole debates with replacement 20,000 times to examine stability. Classify the position argued, not the person's identity or PRO/CON label.",
    "rows": [
      [
        "Comparable religious-versus-skeptical cases",
        "234"
      ],
      [
        "Skeptical higher / religious higher / ties",
        "193 / 33 / 8"
      ],
      [
        "New comparisons / mean gap",
        "47 / 3.34 points"
      ],
      [
        "Mean gap / middle-95% resampling range",
        "5.74 / 4.96–6.52"
      ]
    ],
    "files": [
      "classification.csv",
      "debates.json",
      "results.json",
      "analyze.py"
    ],
    "file": "why-do-the-theist-sides-score-lower",
    "pages": 8,
    "figures": 3,
    "manuscript_words": 1737,
    "path": "output/pdf/why-do-the-theist-sides-score-lower.pdf",
    "sha256": "83d67f9057c4e3569fdbfdc773dfb16b2a401d2e06fbf4149bf3062a840f01d5"
  },
  {
    "id": 2,
    "topic": "Questions and burdens",
    "title": "Where Is the Theist Disadvantage Largest?",
    "short": "Topics and inferential burdens",
    "subtitle": "Stable research categories, more cases, and an uncertain topic order.",
    "stats": [
      [
        "7.28",
        "largest observed topic gap:\nreligion, culture and meaning"
      ],
      [
        "3.04",
        "smallest observed topic gap:\nresurrection and history"
      ],
      [
        "8",
        "stable research groups\nnot the browsing categories"
      ]
    ],
    "finding": "The topic differences remain useful; exact topic rankings remain uncertain.",
    "statistic": "7.28 vs 3.04",
    "statisticLabel": "Mean score gaps: religion, culture and meaning vs resurrection and history",
    "explanation": "Religion, culture and meaning has the largest observed gap across 25 comparisons; resurrection and historical evidence is closer across 24. The former leads only about 32% of repeated-draw calculations, so the apparent ordering should not be treated as settled. All eight observed mean gaps favor the skeptical side.",
    "detail": "The paper keeps eight research groups to make editions comparable, while the website's browsing categories are more detailed. It examines the step from religious usefulness to truth, from an origin cause to a fuller theology, and from sincere testimony to a miracle. Support-score gaps persist in both constructive cases and replies.",
    "limitation": "Topics are exploratory, and the categories mix different speakers, questions and opponents. Overlapping uncertainty makes precise ordering fragile. The eight groups are not the current browsing categories.",
    "pdf": "where-is-the-theist-disadvantage-largest",
    "figure": "p2-moves",
    "links": [
      {
        "id": "loke-linford-physical-reality-cause-beginning-2023",
        "label": "Loke–Linford: origins, causes and agency"
      },
      {
        "id": "horn-paulogia-apostolic-sincerity-2023",
        "label": "Horn–Paulogia: sincerity versus stronger historical claims"
      }
    ],
    "method": "Assign each of the 234 included comparisons once to one of eight research groups. Calculate paired score gaps and within-topic resampling ranges. Move-kind comparisons average qualifying moves within each side before subtracting; both sides must have that kind of move. These are not official weighted overall scores.",
    "rows": [
      [
        "Research comparisons / groups",
        "234 / 8"
      ],
      [
        "Largest observed gap",
        "Religion, culture and meaning: 25 cases · 7.28 points"
      ],
      [
        "Smallest observed gap",
        "Resurrection and history: 24 cases · 3.04 points"
      ],
      [
        "All / high-importance / constructive / reply pairs",
        "234 / 233 / 159 / 233"
      ]
    ],
    "files": [
      "classification.csv",
      "moves.json",
      "results.json",
      "analyze.py"
    ],
    "file": "where-is-the-theist-disadvantage-largest",
    "pages": 9,
    "figures": 3,
    "manuscript_words": 1573,
    "path": "output/pdf/where-is-the-theist-disadvantage-largest.pdf",
    "sha256": "13cf35355a14f739a7ee2edbe64d36f9c9586f633c8b1937dd88c6ff5985ca97"
  },
  {
    "id": 3,
    "topic": "Claims and criticism",
    "title": "Are Theist Arguments More Often Slogan-Like?",
    "short": "Direct evidence and its limits",
    "subtitle": "A dated transcript study, with a separate updated score-based comparison.",
    "stats": [
      [
        "77 / 19",
        "protected uses: religious\nversus skeptical side"
      ],
      [
        "187",
        "historically reviewed transcripts\nSeptember 5 direct study"
      ],
      [
        "234",
        "current score comparisons\nnot new slogan reviews"
      ]
    ],
    "finding": "The direct finding concerns criticism-blocking slogans—and remains a dated study.",
    "statistic": "0.56 vs 0.16",
    "statisticLabel": "Protected uses per 10,000 words · historical 187-transcript review",
    "explanation": "Recalculating the saved direct-review records reproduces 77 religious-side protected uses versus 19 skeptical-side uses. Most reviewed debates—144 of 187—have none detected on either side. The broader unsupported-slogan difference is less secure: its repeated-draw range crosses zero.",
    "detail": "The revised paper separately checks 234 current scorecards: 33.87% versus 13.40% of moves meet a combined low-support, low-clarity and low-care rule, averaging within debates. That warning pattern is not direct slogan detection. No new transcripts were coded for this edition.",
    "limitation": "The direct study used one AI reading per transcript, not independent double coding. Caption errors, missed uses and systematic reader bias remain possible. The newer score-based percentages cannot extend or replace the historical direct counts.",
    "pdf": "are-theist-arguments-more-often-slogan-like",
    "figure": "p3-direct-rates",
    "links": [
      {
        "href": "/insights/data-and-methods/#slogans",
        "label": "Inspect the historical direct review and current score check"
      },
      {
        "search": "Lennox",
        "label": "Explore the Lennox assessments"
      }
    ],
    "method": "For the September direct study, divide detected uses by each side's attributed substantive caption words, multiply by 10,000, then average with equal debate weight. Protected slogans require both substituting for a reason and positively blocking criticism. Recalculate published counts without commissioning new judgments. The separate current warning rule requires Support <70, Clarity <80 and Care <80 on the same scored move.",
    "rows": [
      [
        "Direct review date / transcripts",
        "September 5, 2026 / 187"
      ],
      [
        "Protected counts: religious / skeptical",
        "77 / 19"
      ],
      [
        "Protected rates per 10,000 words",
        "0.56 / 0.16"
      ],
      [
        "Debates with neither side detected",
        "144"
      ],
      [
        "New direct transcript reviews",
        "0"
      ],
      [
        "Current score-warning comparison",
        "234 cases · 20.47 percentage-point difference"
      ]
    ],
    "files": [
      "results.json",
      "analyze.py"
    ],
    "historicalFiles": [
      "light-debates.csv",
      "light-results.json",
      "light-incidents.json",
      "supplementary-results.json",
      "protocol-light.md",
      "editorial-corrections.json"
    ],
    "file": "are-theist-arguments-more-often-slogan-like",
    "pages": 8,
    "figures": 3,
    "manuscript_words": 1752,
    "path": "output/pdf/are-theist-arguments-more-often-slogan-like.pdf",
    "sha256": "3ed3e34588d413f7b9ecfc3ec41894584f69aeabe153659a84dc71e2dd8b9476"
  },
  {
    "id": 4,
    "topic": "Role and responsibility",
    "title": "Does the CON Side Have an Inherent Advantage?",
    "short": "Association is not a built-in bonus",
    "subtitle": "The residual role association is now clearer under one weighting—but still not causal.",
    "stats": [
      [
        "4.44",
        "raw CON-minus-PRO gap\nin comparable debates"
      ],
      [
        "1.33",
        "weighted same-person gap\npositive resampling range"
      ],
      [
        "37",
        "people observed\nin both roles"
      ]
    ],
    "finding": "The weighted same-speaker CON gap is now positive across its displayed range.",
    "statistic": "4.44 → 1.33",
    "statisticLabel": "Raw role gap versus weighted same-speaker comparison · score points",
    "explanation": "CON averages 4.44 points higher across 291 comparable debates. Comparing 37 people who appear in both roles gives a smaller weighted gap of 1.33, with a middle-95% range of 0.32–2.30. Unlike September, this weighted range no longer includes zero.",
    "detail": "The equal-person estimate is 1.09 and its range still narrowly crosses zero. Balancing religious orientations gives 1.73. These calculations answer different questions; their reduction from the raw gap is not a measured percentage of causation.",
    "limitation": "Roles are not randomly assigned. Even the same people face different topics and opponents. A positive weighted range supports investigating a modest residual association, not awarding an automatic bonus or penalty to either role.",
    "pdf": "does-the-con-side-have-an-inherent-advantage",
    "figure": "p4-estimates",
    "links": [
      {
        "id": "rauser-malpass-evidential-problem-evil-2021",
        "label": "Rauser–Malpass: the skeptic's positive evidential burden"
      },
      {
        "href": "/rankings/",
        "label": "Explore individual records"
      }
    ],
    "method": "Calculate CON minus PRO within debates. Compare religious orientations at equal 50/50 weight. For people observed in both roles, compare their role-specific means, first equally and then with weight n(PRO) × n(CON) / total appearances. Resample debates for debate estimates and speakers for same-person estimates.",
    "rows": [
      [
        "Comparable debates",
        "291"
      ],
      [
        "Raw CON gap",
        "4.44"
      ],
      [
        "Same-person equal / weighted",
        "1.09 / 1.33"
      ],
      [
        "Weighted middle-95% range",
        "0.32–2.30"
      ],
      [
        "People observed in both roles",
        "37"
      ]
    ],
    "files": [
      "debates.json",
      "classification.csv",
      "results.json",
      "analyze.py"
    ],
    "file": "does-the-con-side-have-an-inherent-advantage",
    "pages": 7,
    "figures": 2,
    "manuscript_words": 1473,
    "path": "output/pdf/does-the-con-side-have-an-inherent-advantage.pdf",
    "sha256": "d772ea78751f8bb13d896c78a4d08b9c6917ce1ec9d1c0bb03d096c8502d14e6"
  },
  {
    "id": 5,
    "topic": "Beyond labels",
    "title": "Beyond the Fallacy Count",
    "short": "No label is not an acquittal",
    "subtitle": "The overall majority shrinks, while the difference between assessment processes persists.",
    "stats": [
      [
        "54.4%",
        "lower-scoring sides with\nno named-fallacy tag"
      ],
      [
        "17.8%",
        "same rate in the\nlater process"
      ],
      [
        "75",
        "higher-scoring sides\nwith a fallacy tag"
      ]
    ],
    "finding": "Named fallacies neither exhaust argument quality nor determine the result.",
    "statistic": "54.4%",
    "statisticLabel": "Lower-scoring sides without a named-fallacy tag · 161 / 296",
    "explanation": "The combined untagged-loss rate is down from 61.7% in September. It is 80.8% in the earlier process and 17.8% later. The overall percentage depends heavily on the mixture of assessment procedures, so it should not become a universal claim about debate.",
    "detail": "Among 158 comparable untagged losses, 118 trail in five or six scoring areas. Conversely, 75 higher-scoring sides carry a fallacy tag. Read the particular inference and its importance, not just the label count.",
    "limitation": "Tags are accepted annotations, not an exhaustive or independent census of errors. Counts cover assessments rather than independent videos; the shared source for assessments 13 and 125 is disclosed.",
    "pdf": "debates-are-usually-lost-without-a-named-fallacy",
    "figure": "p5-cohorts",
    "links": [
      {
        "href": "/rankings/",
        "label": "Explore category fallacies and biases"
      },
      {
        "href": "/backend/",
        "label": "Read the assessment process"
      }
    ],
    "method": "Exclude 12 ties, identify each lower-scoring side, and count whether it has any public named-fallacy tag. Divide by decisive assessments overall and within each process. Separately compare weighted scoring areas for comparable untagged losses. Count label instances separately from tagged moves.",
    "rows": [
      [
        "Archive: untagged losses",
        "161 / 296 · 54.4%"
      ],
      [
        "Earlier process",
        "139 / 172"
      ],
      [
        "Later process",
        "19 / 107"
      ],
      [
        "Labels / tagged moves",
        "550 / 536"
      ]
    ],
    "files": [
      "losses.json",
      "casebook.json",
      "results.json",
      "analyze.py"
    ],
    "file": "debates-are-usually-lost-without-a-named-fallacy",
    "pages": 8,
    "figures": 3,
    "manuscript_words": 1423,
    "path": "output/pdf/debates-are-usually-lost-without-a-named-fallacy.pdf",
    "sha256": "f3971026a26b57a27076d1dbb1456d75b20b230324741a12c9836024b34f2d4b"
  },
  {
    "id": 6,
    "topic": "Checking the measuring standard",
    "title": "Are All Slugfester Assessments on the Same Scale?",
    "short": "Reconstruction is not calibration",
    "subtitle": "A narrower level difference, persistent process warnings, and a better test for the next review.",
    "stats": [
      [
        "2.35",
        "points lower on average\nin the later process"
      ],
      [
        "56 / 71",
        "returning people with\nlower later means"
      ],
      [
        "7,032",
        "verified comparable\nmove scores"
      ]
    ],
    "finding": "The process-level difference has narrowed, but comparability still needs testing.",
    "statistic": "81.32 vs 78.98",
    "statisticLabel": "Earlier versus later average side scores · 179 / 112 debates",
    "explanation": "The later group's mean is 2.35 points lower, rather than September's 2.82. Of 71 returning people, 56 have lower later averages, 13 higher and 2 unchanged. Annotation rates and the set of exact marks used also differ.",
    "detail": "All 7,032 comparable move scores and 582 overall side scores reconstruct correctly. That verifies the arithmetic, not whether both procedures applied the same judging standard. The next reassessment should test shared source material and publish substantive disagreements.",
    "limitation": "The same people are appearing in different debates, not having the same performance judged twice. Different topics and opponents can contribute. A newer model is not automatically more accurate, and subtracting process means is not a justified public-score correction.",
    "pdf": "are-all-slugfester-assessments-on-the-same-scale",
    "figure": "p6-bridge",
    "links": [
      {
        "href": "/backend/#backend-summary-heading",
        "label": "Read the tentative reassessment plan"
      },
      {
        "href": "/insights/data-and-methods/#same-scale",
        "label": "Inspect the reconstruction and process checks"
      }
    ],
    "method": "Reconstruct all comparable scores from locked records. Compare equal-debate score midpoints, dimension patterns, exact mark frequencies and tag rates by scoring-record format. Follow people appearing in both groups while explicitly retaining topic and opponent limitations.",
    "rows": [
      [
        "Earlier / later debates",
        "179 / 112"
      ],
      [
        "Earlier / later moves",
        "3423 / 3609"
      ],
      [
        "Earlier / later mean side score",
        "81.32 / 78.98"
      ],
      [
        "Returning: lower / higher / unchanged",
        "56 / 13 / 2"
      ]
    ],
    "files": [
      "debates.json",
      "results.json",
      "revision-checks.json",
      "source-manifest.json",
      "analyze.py"
    ],
    "file": "are-all-slugfester-assessments-on-the-same-scale",
    "pages": 8,
    "figures": 4,
    "manuscript_words": 1491,
    "path": "output/pdf/are-all-slugfester-assessments-on-the-same-scale.pdf",
    "sha256": "919022aaa632628963e1d7d4b7f9b201bccc2b850e014dd56b43cad851abd2a7"
  },
  {
    "id": 7,
    "topic": "Reading rankings",
    "title": "Do Slugfester Rankings Measure Stable Performance?",
    "short": "Strong evidence of repeatable performance",
    "subtitle": "Strong repeatability in assessed performance, with less certainty about neighboring places.",
    "stats": [
      [
        "0.87",
        "median order agreement\nbetween random halves"
      ],
      [
        "0.99",
        "order agreement after\nprocess-mean adjustment"
      ],
      [
        "57",
        "people with at least\nthree eligible appearances"
      ]
    ],
    "finding": "Yes—the rankings capture a strong, repeatable pattern of assessed performance.",
    "statistic": "0.87",
    "statisticLabel": "Median ranking agreement between random halves of the record · 1.00 means identical order",
    "explanation": "Among the 31 people with at least six appearances, 3,000 random splits of their debate records produce strongly similar rankings: median correlation 0.87, with the middle 95% of results between 0.81 and 0.92. The broad ordering is therefore not dependent on one particular half of the assessed record. This is strong evidence of stability within the catalogue, not merely a useful way to browse it.",
    "detail": "Two further checks support that conclusion. Dividing records by assessment-number order gives agreement of 0.84; adjusting for average score differences between assessment processes leaves the 57-person ordering almost unchanged (0.99 agreement). The evidence supports broad comparisons of recorded performance, especially for well-represented speakers. Uncertain neighboring places do not negate this stable overall pattern.",
    "limitation": "Stability concerns broad ordering, not every exact place: neighboring means are typically only 0.20 points apart. These correlations are not accuracy percentages or future-win probabilities. The fixed 57-person field contains 416 appearances; topics and opponents are unequal, and repeatability alone cannot establish that the judging is correct.",
    "pdf": "do-slugfester-rankings-measure-stable-performance",
    "figure": "p7-ranges",
    "links": [
      {
        "href": "/interlocutor/joseph-schmid/",
        "label": "Joseph Schmid's current record"
      },
      {
        "href": "/interlocutor/susan-blackmore/",
        "label": "Susan Blackmore's current record"
      },
      {
        "href": "/rankings/",
        "label": "Open current rankings"
      }
    ],
    "method": "Measure Spearman rank correlation between two halves of each person's record in 3,000 random splits, restricted to the 31 people with at least six appearances. The 0.81–0.92 interval covers the middle 95% of split results, not a population confidence interval. Check assessment-number splits (not necessarily recording chronology) and process-mean adjustment separately. For the full 57-person field with at least three appearances, compare 20,000 resampled ranks with a normal model using pooled within-person variation. Shared debate weights preserve opponent pairing. These complementary checks reuse the same archive; they are not independent replications.",
    "rows": [
      [
        "Ranked people / appearances",
        "57 / 416"
      ],
      [
        "All eligible people / appearances",
        "183 / 582"
      ],
      [
        "Random-half agreement / middle 95% of splits (31 people)",
        "0.87 / 0.81–0.92"
      ],
      [
        "Assessment-number split agreement (31 people)",
        "0.84"
      ],
      [
        "Process-adjusted order agreement (57 people)",
        "0.99"
      ],
      [
        "Modeled repeatability: single / three / ten appearances",
        "0.61 / 0.83 / 0.94"
      ],
      [
        "Typical neighboring gap",
        "0.20 points"
      ],
      [
        "Typical resampled / model rank width",
        "14 / 21 places"
      ]
    ],
    "files": [
      "ranking.csv",
      "results.json",
      "analyze.py"
    ],
    "file": "do-slugfester-rankings-measure-stable-performance",
    "pages": 12,
    "figures": 3,
    "manuscript_words": 1918,
    "path": "output/pdf/do-slugfester-rankings-measure-stable-performance.pdf",
    "sha256": "f7c8a21bf970b71933ff370a3e09e9dec899d999683f301ba264bf1d037fd0db"
  }
]