{
  "schema_version": "1.0.0",
  "title": "Convexly V1-M sweepcash politics-excluded sensitivity rerun",
  "generated_date_utc": "2026-05-04",
  "status": "correction packet for the public V1-M methodology paper; not a byte-identical rebuild of the 2026-04-22 bundle",
  "source_question": "Does the V1-M within-user concentration result survive when political and election markets are excluded, as requested by Tod Waddington on 2026-04-30?",
  "public_v1m_bundle": {
    "published_date": "2026-04-22",
    "comparable_pre_to_sweep_concentration_delta": -0.0892,
    "comparable_pre_to_sweep_paired_users": 1647,
    "comparable_sweep_to_post_concentration_delta": 0.0354,
    "comparable_sweep_to_post_paired_users": 1839,
    "caveat": "The original public headline_numbers.json reported point estimates only. This artifact adds recovered-cohort inferential statistics and a politics-excluded sensitivity arm."
  },
  "recovered_cohort": {
    "public_v1m_user_hashes": 15106,
    "recovered_user_ids": 15027,
    "unrecovered_public_hashes": 79,
    "recovered_share": 0.9948,
    "byte_identical_rebuild": false,
    "record_hash_validation": {
      "sweepcash_baseline_records.jsonl": {
        "records": 3205,
        "unique_hashes": 3205,
        "missing_from_public_v1m_user_hashes": 0
      },
      "sweepcash_politics_excluded_records.jsonl": {
        "records": 1789,
        "unique_hashes": 1789,
        "missing_from_public_v1m_user_hashes": 0
      }
    }
  },
  "politics_filter": {
    "applied": true,
    "tier_cutoff": 4,
    "excluded_contract_count": 28999,
    "tier_histogram": {
      "non_politics": 137298,
      "tier1": 106,
      "tier2": 18366,
      "tier3": 0,
      "tier4": 10527
    },
    "source_taxonomy": "Tod Waddington email 2026-05-01 + github.com/manifoldmarkets/manifold (elections-data + senate-state + governors + politics/home + topics.ts)"
  },
  "headline": {
    "baseline_no_politics_filter": {
      "cohort_label": "baseline",
      "transition": "gap_pre_sweepcash -> sweepcash",
      "metric": "concentration",
      "n_paired_users": 2330,
      "n_with_delta_defined": 497,
      "median_delta": -0.10628,
      "ci_95_bootstrap": [
        -0.16146,
        -0.01388
      ],
      "boot_se": 0.03834,
      "wilcoxon_p_value": 0.0079255905910449,
      "wilcoxon_n_nonzero": 497,
      "ci_excludes_zero": true
    },
    "politics_excluded": {
      "cohort_label": "politics_excluded",
      "transition": "gap_pre_sweepcash -> sweepcash",
      "metric": "concentration",
      "n_paired_users": 1208,
      "n_with_delta_defined": 515,
      "median_delta": -0.09184,
      "ci_95_bootstrap": [
        -0.12804,
        -0.03627
      ],
      "boot_se": 0.02332,
      "wilcoxon_p_value": 0.002116676567018907,
      "wilcoxon_n_nonzero": 515,
      "ci_excludes_zero": true
    },
    "comparison": {
      "ci_excludes_zero_baseline": true,
      "ci_excludes_zero_filtered": true,
      "direction_preserved_after_filter": true,
      "magnitude_shift_pct_filtered_vs_baseline": -13.6,
      "direction_word": "attenuating",
      "cohort_retention_pct_filtered_vs_baseline": 103.6,
      "baseline_filtered_cis_overlap": true,
      "inferential_conclusion_holds_after_filter": true
    },
    "interpretation": "The politics filter attenuates the recovered-cohort concentration result by 13.6%, but the politics-excluded confidence interval still excludes zero."
  },
  "calibration_correction": {
    "baseline_no_politics_filter": {
      "cohort_label": "baseline",
      "transition": "gap_pre_sweepcash -> sweepcash",
      "metric": "skill_brier",
      "n_paired_users": 2330,
      "n_with_delta_defined": 2330,
      "median_delta": 0.00046,
      "ci_95_bootstrap": [
        -0.00252,
        0.00356
      ],
      "boot_se": 0.00154,
      "wilcoxon_p_value": 0.9319943387501405,
      "wilcoxon_n_nonzero": 2330,
      "ci_excludes_zero": false,
      "tost_equivalence": {
        "+/- 0.005": {
          "n": 2330,
          "mean": -0.00136,
          "ci_lo_90_one_sided": -0.00405,
          "ci_hi_90_one_sided": 0.00128,
          "equivalence_margin": 0.005,
          "h0_lower_rejected": true,
          "h0_upper_rejected": true,
          "equivalent_within_margin": true
        },
        "+/- 0.01": {
          "n": 2330,
          "mean": -0.00136,
          "ci_lo_90_one_sided": -0.00403,
          "ci_hi_90_one_sided": 0.00129,
          "equivalence_margin": 0.01,
          "h0_lower_rejected": true,
          "h0_upper_rejected": true,
          "equivalent_within_margin": true
        }
      }
    },
    "politics_excluded": {
      "cohort_label": "politics_excluded",
      "transition": "gap_pre_sweepcash -> sweepcash",
      "metric": "skill_brier",
      "n_paired_users": 1208,
      "n_with_delta_defined": 1208,
      "median_delta": -0.00084,
      "ci_95_bootstrap": [
        -0.00488,
        0.00293
      ],
      "boot_se": 0.00199,
      "wilcoxon_p_value": 0.9435991220445094,
      "wilcoxon_n_nonzero": 1208,
      "ci_excludes_zero": false,
      "tost_equivalence": {
        "+/- 0.005": {
          "n": 1208,
          "mean": -0.0002,
          "ci_lo_90_one_sided": -0.00418,
          "ci_hi_90_one_sided": 0.00374,
          "equivalence_margin": 0.005,
          "h0_lower_rejected": true,
          "h0_upper_rejected": true,
          "equivalent_within_margin": true
        },
        "+/- 0.01": {
          "n": 1208,
          "mean": -0.0002,
          "ci_lo_90_one_sided": -0.0042,
          "ci_hi_90_one_sided": 0.0039,
          "equivalence_margin": 0.01,
          "h0_lower_rejected": true,
          "h0_upper_rejected": true,
          "equivalent_within_margin": true
        }
      }
    },
    "interpretation": "The original calibration-improvement secondary claim does not replicate under the recovered cohort or the politics-excluded sensitivity arm."
  },
  "sweep_to_post_context": {
    "concentration": {
      "baseline": {
        "cohort_label": "baseline",
        "transition": "sweepcash -> post_sweepcash",
        "metric": "concentration",
        "n_paired_users": 2137,
        "n_with_delta_defined": 495,
        "median_delta": 0.08061,
        "ci_95_bootstrap": [
          0.01495,
          0.15286
        ],
        "boot_se": 0.03712,
        "wilcoxon_p_value": 0.016845315313829563,
        "wilcoxon_n_nonzero": 495,
        "ci_excludes_zero": true
      },
      "filtered": {
        "cohort_label": "politics_excluded",
        "transition": "sweepcash -> post_sweepcash",
        "metric": "concentration",
        "n_paired_users": 1284,
        "n_with_delta_defined": 545,
        "median_delta": 0.0303,
        "ci_95_bootstrap": [
          -0.00199,
          0.07305
        ],
        "boot_se": 0.02096,
        "wilcoxon_p_value": 0.056179152392419884,
        "wilcoxon_n_nonzero": 545,
        "ci_excludes_zero": false
      },
      "comparison": {
        "ci_excludes_zero_baseline": true,
        "ci_excludes_zero_filtered": false,
        "direction_preserved_after_filter": true,
        "magnitude_shift_pct_filtered_vs_baseline": -62.4,
        "direction_word": "attenuating",
        "cohort_retention_pct_filtered_vs_baseline": 110.1,
        "baseline_filtered_cis_overlap": true,
        "inferential_conclusion_holds_after_filter": false
      }
    },
    "skill_brier": {
      "baseline": {
        "cohort_label": "baseline",
        "transition": "sweepcash -> post_sweepcash",
        "metric": "skill_brier",
        "n_paired_users": 2137,
        "n_with_delta_defined": 2137,
        "median_delta": 0.01243,
        "ci_95_bootstrap": [
          0.00963,
          0.0151
        ],
        "boot_se": 0.00138,
        "wilcoxon_p_value": 3.116046819292552e-19,
        "wilcoxon_n_nonzero": 2137,
        "ci_excludes_zero": true,
        "tost_equivalence": {
          "+/- 0.005": {
            "n": 2137,
            "mean": 0.01156,
            "ci_lo_90_one_sided": 0.00876,
            "ci_hi_90_one_sided": 0.01443,
            "equivalence_margin": 0.005,
            "h0_lower_rejected": true,
            "h0_upper_rejected": false,
            "equivalent_within_margin": false
          },
          "+/- 0.01": {
            "n": 2137,
            "mean": 0.01156,
            "ci_lo_90_one_sided": 0.00869,
            "ci_hi_90_one_sided": 0.01437,
            "equivalence_margin": 0.01,
            "h0_lower_rejected": true,
            "h0_upper_rejected": false,
            "equivalent_within_margin": false
          }
        }
      },
      "filtered": {
        "cohort_label": "politics_excluded",
        "transition": "sweepcash -> post_sweepcash",
        "metric": "skill_brier",
        "n_paired_users": 1284,
        "n_with_delta_defined": 1284,
        "median_delta": 0.0134,
        "ci_95_bootstrap": [
          0.00975,
          0.0189
        ],
        "boot_se": 0.00237,
        "wilcoxon_p_value": 3.700750218032195e-12,
        "wilcoxon_n_nonzero": 1284,
        "ci_excludes_zero": true,
        "tost_equivalence": {
          "+/- 0.005": {
            "n": 1284,
            "mean": 0.01245,
            "ci_lo_90_one_sided": 0.00859,
            "ci_hi_90_one_sided": 0.0163,
            "equivalence_margin": 0.005,
            "h0_lower_rejected": true,
            "h0_upper_rejected": false,
            "equivalent_within_margin": false
          },
          "+/- 0.01": {
            "n": 1284,
            "mean": 0.01245,
            "ci_lo_90_one_sided": 0.00857,
            "ci_hi_90_one_sided": 0.01625,
            "equivalence_margin": 0.01,
            "h0_lower_rejected": true,
            "h0_upper_rejected": false,
            "equivalent_within_margin": false
          }
        }
      },
      "comparison": {
        "ci_excludes_zero_baseline": true,
        "ci_excludes_zero_filtered": true,
        "direction_preserved_after_filter": true,
        "magnitude_shift_pct_filtered_vs_baseline": 7.8,
        "direction_word": "unchanged",
        "cohort_retention_pct_filtered_vs_baseline": 60.1,
        "baseline_filtered_cis_overlap": true,
        "inferential_conclusion_holds_after_filter": true
      }
    },
    "interpretation": "The concentration reversal is filter-sensitive and should be framed as suggestive, not settled. The sweep-to-post skill_brier worsening also survives the politics filter."
  },
  "guardrails": [
    "Use within-user paired comparison across the sweepcash window, not randomized intervention.",
    "Avoid natural experiment, proves, real money produced, and pre-registered V1-M language.",
    "Disclose both paired-user count and defined-delta sub-sample count in the same sentence.",
    "State that V1 and V1-M were not externally pre-registered; V1.5 follow-up tests were externally pre-registered at AsPredicted #287368."
  ],
  "public_note_for_tod": "Recovered-cohort rerun: 15,027 of 15,106 public V1-M user hashes recovered (99.5%; 79 unrecovered). This is NOT a byte-identical rebuild of the 2026-04-22 public bundle. Headline within-user concentration delta across the pre-sweepcash to sweepcash window transition: baseline -10.6pp, 95% bootstrap CI [-16.1, -1.4], Wilcoxon p=0.0079 on n=497 of 2,330 paired users with concentration delta defined (concentration is undefined when total realized PnL is non-positive in either window); politics-excluded sensitivity rerun -9.2pp, 95% bootstrap CI [-12.8, -3.6], Wilcoxon p=0.0021 on n=515 of 1,208 paired users (defined-delta cohort retention 103.6% vs baseline). Direction word filtered-vs-baseline: **attenuating** (magnitude shift -13.6% of baseline magnitude). The CI-excludes-zero conclusion still holds after the politics filter. Framing: this is a within-user paired comparison across two windows, not a randomized intervention; causal attribution to the real-money sweepcash switch is not a clean inference from this design. The published calibration (skill_brier) secondary claim does not replicate in the recovered-cohort rerun, and the post-sweepcash symmetric-reversal concentration result is filter-sensitive; both are reported in the per_window_transition block and any external send should disclose them alongside the headline."
}