{
  "trained_at": "2026-04-27T12:00:40.349432+00:00",
  "data_run_date": "2026-04-25",
  "llm_run_date": "2026-04-26",
  "version": "v0.8-fixed-wash-ensemble",
  "description": "v0.7 features + tightened wash detection + dual-tree ensemble + bias correction",
  "parent_version": "v0.7",
  "changes_from_parent": [
    "wash detection: row-level, \u226424h + \u22651.5x price gain",
    "wash detection: dropped wallet concentration heuristic",
    "dual-tree ensemble: tuned XGBoost + tuned LightGBM, log-space mean",
    "val-derived bias correction (val_bias = -0.0781)",
    "4x Optuna budget (120 trials per model)"
  ],
  "splits": {
    "train": {
      "rows": 265240,
      "start": "2022-01-28",
      "end": "2023-09-30"
    },
    "val": {
      "rows": 3545,
      "start": "2023-10-01",
      "end": "2023-12-31"
    },
    "test": {
      "rows": 2744,
      "start": "2024-01-01",
      "end": "2024-05-04"
    }
  },
  "feature_count": 212,
  "feature_cols": [
    "pca_000",
    "pca_001",
    "pca_002",
    "pca_003",
    "pca_004",
    "pca_005",
    "pca_006",
    "pca_007",
    "pca_008",
    "pca_009",
    "pca_010",
    "pca_011",
    "pca_012",
    "pca_013",
    "pca_014",
    "pca_015",
    "pca_016",
    "pca_017",
    "pca_018",
    "pca_019",
    "pca_020",
    "pca_021",
    "pca_022",
    "pca_023",
    "pca_024",
    "pca_025",
    "pca_026",
    "pca_027",
    "pca_028",
    "pca_029",
    "pca_030",
    "pca_031",
    "pca_032",
    "pca_033",
    "pca_034",
    "pca_035",
    "pca_036",
    "pca_037",
    "pca_038",
    "pca_039",
    "pca_040",
    "pca_041",
    "pca_042",
    "pca_043",
    "pca_044",
    "pca_045",
    "pca_046",
    "pca_047",
    "pca_048",
    "pca_049",
    "pca_050",
    "pca_051",
    "pca_052",
    "pca_053",
    "pca_054",
    "pca_055",
    "pca_056",
    "pca_057",
    "pca_058",
    "pca_059",
    "pca_060",
    "pca_061",
    "pca_062",
    "pca_063",
    "len",
    "n_digits",
    "n_letters",
    "n_special",
    "n_lower",
    "n_upper",
    "is_palindrome",
    "is_all_digits",
    "is_all_letters",
    "is_ascii",
    "has_unicode",
    "starts_digit",
    "ends_digit",
    "max_char_run",
    "n_unique_chars",
    "in_wikipedia",
    "in_geonames",
    "in_us_firstname",
    "in_iso3166",
    "in_ticker",
    "in_sec_edgar",
    "in_wiktionary_en",
    "wordlist_hits",
    "club__prepunk_full_rankings",
    "club__firstnames_usa",
    "club__finance_terms",
    "club__social_handles",
    "club__gamertags",
    "club__top_crypto_tickers",
    "club__common_english",
    "club__sports",
    "club__crypto_terms",
    "club__gamertags_double",
    "club__pokemon_gen4",
    "club__fine_art",
    "club__logistics",
    "club__top500_cities_usa",
    "club__natural_wonders",
    "club__familynames_usa",
    "club__top_nouns",
    "club__wikidata_top_fantasy_char",
    "club__holidays",
    "club__us_government",
    "club__top500_cities_global",
    "club__pokemon_gen3",
    "club__common_animals",
    "club__country_codes",
    "club__personas",
    "club__currency_symbols",
    "club__mythical_creatures",
    "club__pokemon_gen1",
    "club__catholicism",
    "club__historic_figures",
    "club__luxury",
    "club__paranormal",
    "club__crayola_classic",
    "club__top_crypto_names",
    "club__performing_arts",
    "club__pokemon_gen2",
    "club__home",
    "club__conspiracy_theories",
    "club__memes",
    "club__us_states",
    "club__currency_names",
    "club__gen_alpha",
    "n_clubs",
    "trademark_conflict",
    "name_age_days",
    "prior_transfer_count",
    "fg_value",
    "eth_tvl_usd",
    "eth_stable_mcap",
    "eth_dex_volume",
    "nft_total_fee_usd",
    "fame_score",
    "crypto_relevance_ord",
    "brand_collision_risk_ord",
    "kind__concept",
    "kind__random",
    "kind__brand",
    "kind__surname",
    "kind__first_name",
    "kind__abbreviation",
    "kind__place",
    "kind__other",
    "kind__unknown",
    "origin__english",
    "origin__none",
    "origin__mixed",
    "origin__spanish",
    "origin__german",
    "origin__french",
    "origin__japanese",
    "origin__chinese",
    "origin__italian",
    "origin__slavic",
    "origin__korean",
    "origin__arabic",
    "origin__other",
    "origin__unknown",
    "desc_pca_000",
    "desc_pca_001",
    "desc_pca_002",
    "desc_pca_003",
    "desc_pca_004",
    "desc_pca_005",
    "desc_pca_006",
    "desc_pca_007",
    "desc_pca_008",
    "desc_pca_009",
    "desc_pca_010",
    "desc_pca_011",
    "desc_pca_012",
    "desc_pca_013",
    "desc_pca_014",
    "desc_pca_015",
    "desc_pca_016",
    "desc_pca_017",
    "desc_pca_018",
    "desc_pca_019",
    "desc_pca_020",
    "desc_pca_021",
    "desc_pca_022",
    "desc_pca_023",
    "desc_pca_024",
    "desc_pca_025",
    "desc_pca_026",
    "desc_pca_027",
    "desc_pca_028",
    "desc_pca_029",
    "desc_pca_030",
    "desc_pca_031",
    "knnmp_count",
    "knnmp_mean_log",
    "knnmp_median_log",
    "knnmp_p90_log",
    "knnmp_max_sim",
    "knnmp_min_sim",
    "knnmp_log_max",
    "knnmp_log_min",
    "knnbg_count",
    "knnbg_mean_log",
    "knnbg_median_log",
    "knnbg_p90_log",
    "knnbg_max_sim",
    "knnbg_min_sim",
    "knnbg_log_max",
    "knnbg_log_min"
  ],
  "pca_dim_concat": 64,
  "pca_dim_description": 32,
  "name_kind_values": [
    "concept",
    "random",
    "brand",
    "surname",
    "first_name",
    "abbreviation",
    "place",
    "other",
    "unknown"
  ],
  "cultural_origin_values": [
    "english",
    "none",
    "mixed",
    "spanish",
    "german",
    "french",
    "japanese",
    "chinese",
    "italian",
    "slavic",
    "korean",
    "arabic",
    "other",
    "unknown"
  ],
  "wash_filtering": {
    "method": "row_level",
    "round_trip_max_hours": 24,
    "round_trip_min_price_gain": 1.5,
    "wash_train_weight": 0.1,
    "n_round_trip_rows": 8236,
    "n_sub_minute_rows": 2612,
    "n_total_flagged_rows": 10733,
    "flag_rate_pct": 3.9528006216647213
  },
  "optuna": {
    "n_trials_per_model": 120,
    "xgb_best_val_rmse": 1.0110926628112793,
    "xgb_best_params": {
      "max_depth": 12,
      "learning_rate": 0.00707952961758631,
      "subsample": 0.82649559352978,
      "colsample_bytree": 0.5076045191428247,
      "colsample_bylevel": 0.4490941323703761,
      "min_child_weight": 25,
      "reg_alpha": 2.1100924086636215,
      "reg_lambda": 6.7085987791616875,
      "gamma": 3.012785755618491
    },
    "xgb_best_trial": 78,
    "lgb_best_val_rmse": 1.0083379004456063,
    "lgb_best_params": {
      "boosting_type": "gbdt",
      "num_leaves": 294,
      "max_depth": 13,
      "learning_rate": 0.007555121763111837,
      "feature_fraction": 0.714483668555124,
      "feature_fraction_bynode": 0.5304393257914843,
      "min_data_in_leaf": 6,
      "lambda_l1": 0.002471118357836334,
      "lambda_l2": 1.776020947969551,
      "min_gain_to_split": 2.2135701585440897,
      "bagging_fraction": 0.9999413018023948,
      "bagging_freq": 7
    },
    "lgb_best_trial": 16
  },
  "xgb_quantile_models": {
    "q05": {
      "best_iteration": 984,
      "best_val_rmse": 1.9671143709945549
    },
    "q50": {
      "best_iteration": 1756,
      "best_val_rmse": 1.0471049311283205
    },
    "q95": {
      "best_iteration": 3964,
      "best_val_rmse": 2.1785484376377413
    }
  },
  "lgb_quantile_models": {
    "q05": {
      "best_iteration": 500
    },
    "q50": {
      "best_iteration": 1465
    },
    "q95": {
      "best_iteration": 1123
    }
  },
  "ensemble": {
    "method": "log_space_mean",
    "members": [
      "xgb_q50",
      "lgb_q50"
    ]
  },
  "bias_correction": {
    "method": "val_derived_constant",
    "val_bias": -0.0781494239191985
  },
  "calibration": {
    "method": "additive",
    "target_coverage": 0.9,
    "delta_lower": 0.14621443811853208,
    "delta_upper": 0.0,
    "val_bias_correction": -0.0781494239191985,
    "ensemble_method": "log_space_mean_xgb_lgb"
  },
  "metrics": {
    "final": {
      "train": {
        "r2_log": 0.7860012517591471,
        "rmse_log": 0.7469926912886737,
        "mae_log": 0.45205338808523504,
        "median_ape": 0.24072927576014078,
        "bias_log": -0.0053798850758574246
      },
      "val": {
        "r2_log": 0.6758550101826879,
        "rmse_log": 1.0429795602982888,
        "mae_log": 0.706429696381903,
        "median_ape": 0.43773059069184295,
        "bias_log": -1.8039166775291682e-16
      },
      "test": {
        "r2_log": 0.47530460716475276,
        "rmse_log": 1.3470754265844747,
        "mae_log": 1.0706065416171697,
        "median_ape": 0.9519209571310361,
        "bias_log": 0.444023565817704
      }
    },
    "coverage": {
      "train": {
        "coverage_90pct": 0.937335997587091,
        "median_interval_log": 2.823927108304763,
        "median_interval_ratio": 16.84286471905812
      },
      "val": {
        "coverage_90pct": 0.8493653032440056,
        "median_interval_log": 3.6384351371935924,
        "median_interval_ratio": 38.03227484212217
      },
      "test": {
        "coverage_90pct": 0.8126822157434402,
        "median_interval_log": 4.143242668045713,
        "median_interval_ratio": 63.00680069124568
      }
    }
  },
  "top_features_xgb": [
    {
      "name": "knnmp_mean_log",
      "gain": 108.798828125
    },
    {
      "name": "knnmp_median_log",
      "gain": 78.58806610107422
    },
    {
      "name": "knnmp_log_min",
      "gain": 72.71774291992188
    },
    {
      "name": "len",
      "gain": 61.729530334472656
    },
    {
      "name": "knnmp_p90_log",
      "gain": 58.61769104003906
    },
    {
      "name": "pca_004",
      "gain": 42.28458786010742
    },
    {
      "name": "knnmp_log_max",
      "gain": 39.73566818237305
    },
    {
      "name": "is_all_digits",
      "gain": 38.88374710083008
    },
    {
      "name": "knnmp_count",
      "gain": 36.54442596435547
    },
    {
      "name": "pca_002",
      "gain": 36.348838806152344
    },
    {
      "name": "kind__random",
      "gain": 34.42299270629883
    },
    {
      "name": "in_wikipedia",
      "gain": 28.338823318481445
    },
    {
      "name": "eth_stable_mcap",
      "gain": 28.019874572753906
    },
    {
      "name": "origin__none",
      "gain": 27.93862533569336
    },
    {
      "name": "n_digits",
      "gain": 26.823389053344727
    },
    {
      "name": "ends_digit",
      "gain": 25.347196578979492
    },
    {
      "name": "n_unique_chars",
      "gain": 25.27269172668457
    },
    {
      "name": "name_age_days",
      "gain": 25.112060546875
    },
    {
      "name": "eth_tvl_usd",
      "gain": 24.409042358398438
    },
    {
      "name": "starts_digit",
      "gain": 23.97994613647461
    },
    {
      "name": "knnbg_count",
      "gain": 23.276514053344727
    },
    {
      "name": "is_palindrome",
      "gain": 22.497846603393555
    },
    {
      "name": "brand_collision_risk_ord",
      "gain": 21.805770874023438
    },
    {
      "name": "origin__chinese",
      "gain": 21.738651275634766
    },
    {
      "name": "n_letters",
      "gain": 20.87726402282715
    },
    {
      "name": "kind__abbreviation",
      "gain": 20.590986251831055
    },
    {
      "name": "pca_001",
      "gain": 20.483781814575195
    },
    {
      "name": "has_unicode",
      "gain": 19.911903381347656
    },
    {
      "name": "nft_total_fee_usd",
      "gain": 19.79189682006836
    },
    {
      "name": "desc_pca_000",
      "gain": 19.757749557495117
    }
  ],
  "top_features_lgb": [
    {
      "name": "knnmp_mean_log",
      "gain": 781449.8409805298
    },
    {
      "name": "eth_stable_mcap",
      "gain": 780319.1262354851
    },
    {
      "name": "name_age_days",
      "gain": 621649.1777780056
    },
    {
      "name": "eth_tvl_usd",
      "gain": 549847.3152387142
    },
    {
      "name": "len",
      "gain": 535403.315526247
    },
    {
      "name": "knnmp_median_log",
      "gain": 476187.2463479042
    },
    {
      "name": "nft_total_fee_usd",
      "gain": 376482.9395339489
    },
    {
      "name": "knnmp_p90_log",
      "gain": 299560.9693763256
    },
    {
      "name": "fg_value",
      "gain": 291222.41970157623
    },
    {
      "name": "n_digits",
      "gain": 200339.19226312637
    },
    {
      "name": "knnmp_count",
      "gain": 198496.79864764214
    },
    {
      "name": "knnmp_log_min",
      "gain": 196220.42209792137
    },
    {
      "name": "eth_dex_volume",
      "gain": 183750.19882249832
    },
    {
      "name": "n_unique_chars",
      "gain": 178599.7514090538
    },
    {
      "name": "knnmp_log_max",
      "gain": 165553.80459403992
    },
    {
      "name": "pca_004",
      "gain": 119861.93015813828
    },
    {
      "name": "desc_pca_000",
      "gain": 114556.08047294617
    },
    {
      "name": "pca_002",
      "gain": 104693.19005703926
    },
    {
      "name": "n_letters",
      "gain": 83801.49758315086
    },
    {
      "name": "pca_001",
      "gain": 82503.51477956772
    },
    {
      "name": "n_special",
      "gain": 70174.17102837563
    },
    {
      "name": "is_palindrome",
      "gain": 64647.565772771835
    },
    {
      "name": "is_all_digits",
      "gain": 56822.11631703377
    },
    {
      "name": "desc_pca_002",
      "gain": 47493.986828804016
    },
    {
      "name": "n_lower",
      "gain": 44417.38100814819
    },
    {
      "name": "max_char_run",
      "gain": 40815.91745662689
    },
    {
      "name": "knnbg_count",
      "gain": 39852.69767045975
    },
    {
      "name": "fame_score",
      "gain": 36314.24634027481
    },
    {
      "name": "pca_005",
      "gain": 34642.322847127914
    },
    {
      "name": "knnmp_min_sim",
      "gain": 34452.339801073074
    }
  ],
  "family_gain_split_xgb": {
    "mpnet_ft_knn": 416.6655960083008,
    "bge_knn": 94.33125257492065,
    "llm_kind": 89.02087259292603,
    "llm_origin": 129.38252782821655,
    "llm_scores": 46.957244873046875,
    "llm_desc": 279.1159596443176,
    "llm_total": 544.4766049385071
  },
  "inference_recipe": {
    "description": "Inference uses 6 models (XGB\u00d73 + LGB\u00d73) + bias + calibration constants",
    "point_estimate": "(xgb_q50.predict() + lgb_q50.predict()) / 2 - val_bias",
    "uncalibrated_low": "(xgb_q05.predict() + lgb_q05.predict()) / 2 - val_bias",
    "uncalibrated_high": "(xgb_q95.predict() + lgb_q95.predict()) / 2 - val_bias",
    "calibrated_low": "uncalibrated_low - delta_lower",
    "calibrated_high": "uncalibrated_high + delta_upper",
    "output_usd": "np.exp(final)"
  },
  "wandb_run": "https://wandb.ai/quantumly-aletheia-research/ens-appraiser/runs/d3y6y5c6"
}