{
 "bootstrap": {
  "n": 2000,
  "seed": 0,
  "unit": "recording"
 },
 "code_dirty": [
  "hf-release-v0.1/config.json",
  "training/hf_convert.py",
  "training/hq5a_cough_eval.sbatch",
  "training/hq5a_evalsets.sbatch",
  "training/release_gate.py",
  "hf-release-v0.1/release-manifest.json",
  "training/hq5a_event_calib.sbatch"
 ],
 "code_revision": "e4db7af66b151e230c6075e28253302b9b48ff6e",
 "family": "laughter-v0.1",
 "members": [
  "laughter-v0.1-seed0",
  "laughter-v0.1-seed1",
  "laughter-v0.1-seed10",
  "laughter-v0.1-seed11",
  "laughter-v0.1-seed12",
  "laughter-v0.1-seed13",
  "laughter-v0.1-seed14",
  "laughter-v0.1-seed2",
  "laughter-v0.1-seed3",
  "laughter-v0.1-seed4",
  "laughter-v0.1-seed5",
  "laughter-v0.1-seed6",
  "laughter-v0.1-seed7",
  "laughter-v0.1-seed8",
  "laughter-v0.1-seed9"
 ],
 "reproduce": "For each public surface: AP of each probs/<surface>__laughter-v0.1-seedN.npy against target>=0.5 in <surface>.clips.tsv, mean over the 15 seeds = family_mean; margin = family_mean - reference AP; CI by paired recording (speaker on VocalSound) bootstrap, 2000 resamples, seed 0.",
 "rule": "internal-and-any-public",
 "set_digests": {
  "ami-eval-v1": "1ab59b11012d18d9",
  "annotation-eval-v5": "2dcdba95c052bbde",
  "podcastfillers-eval": "e4deca38c6380c35",
  "public-vocalsound": "c4557271c7f68f70",
  "fsd50k-panel": "b55cd8789c24fda8"
 },
 "table": {
  "ami-eval-v1": {
   "ci_hi": 0.03942520109963582,
   "ci_lo": 0.012154065156311903,
   "family_mean": 0.9483739882117678,
   "floor_ci_upper": 0.0039598598827272615,
   "floor_sigma": 0.002556051282820607,
   "label": "AMI",
   "margin": 0.02538138623377517,
   "member_ap": [
    0.9507902186015077,
    0.9500519992061698,
    0.9452210675451327,
    0.9503056961627866,
    0.9508894430969833,
    0.9470331036343588,
    0.9450138166749027,
    0.9465480039960334,
    0.9513989118555961,
    0.9483455133944823,
    0.9432426745514125,
    0.9511137070448046,
    0.9483085823703894,
    0.9485920621237138,
    0.9487550229182443
   ],
   "n": 15,
   "n_clips": 3082,
   "n_recordings": 24,
   "reference_ap": 0.9229926019779926,
   "reference_model": "eat"
  },
  "annotation-eval-v5": {
   "ci_hi": 0.023578607063105828,
   "ci_lo": -0.014470678820449834,
   "family_mean": 0.8996943593326668,
   "floor_ci_upper": 0.004720526387522345,
   "floor_sigma": 0.003047054159932776,
   "label": "internal",
   "margin": 0.0028852774625475197,
   "member_ap": [
    0.8975475415661825,
    0.9039947028281456,
    0.8983932783475506,
    0.8993346068750109,
    0.9005183386530095,
    0.9016603462847035,
    0.9028680945100267,
    0.8955774413407963,
    0.9002877424905875,
    0.9031921793787044,
    0.8952398854096171,
    0.9011704571931498,
    0.8995806005577214,
    0.8942340859359875,
    0.9018160886188078
   ],
   "n": 15,
   "n_clips": 2297,
   "n_recordings": 21,
   "reference_ap": 0.8968090818701193,
   "reference_model": "ced-base",
   "_per_clip": "not shipped (internal audio)"
  },
  "podcastfillers-eval": {
   "ci_hi": 0.050521526766975876,
   "ci_lo": -0.02075922278013193,
   "family_mean": 0.8333527146573345,
   "floor_ci_upper": 0.01673321979650875,
   "floor_sigma": 0.010801131654468506,
   "label": "PF",
   "margin": 0.014778337531855201,
   "member_ap": [
    0.8344361951641851,
    0.841598928384278,
    0.8411140826012734,
    0.8366294402860901,
    0.8337691285784827,
    0.8382277485746962,
    0.8403275294236768,
    0.831845088772049,
    0.8383355800012807,
    0.8350686755602265,
    0.7975593280141196,
    0.8376021852481,
    0.8343551388336643,
    0.8279070269767151,
    0.8315146434411776
   ],
   "n": 15,
   "n_clips": 10398,
   "n_recordings": 20,
   "reference_ap": 0.8185743771254793,
   "reference_model": "gillick"
  },
  "public-vocalsound": {
   "ci_hi": -0.005603668917087099,
   "ci_lo": -0.02523479335876062,
   "family_mean": 0.9550608658482184,
   "floor_ci_upper": 0.007326952674220729,
   "floor_sigma": 0.004729477137259021,
   "label": "VS",
   "margin": -0.015235673719981468,
   "member_ap": [
    0.9520376579742674,
    0.9552183790146299,
    0.9540970381377161,
    0.9509495673970976,
    0.9519229996151687,
    0.9537347287207015,
    0.9557549955025102,
    0.9619265200931599,
    0.9544285936941963,
    0.9517720709172127,
    0.9652754551778516,
    0.9524594157733894,
    0.9488156226153927,
    0.9624520243219058,
    0.9550679187680756
   ],
   "n": 15,
   "n_clips": 3591,
   "n_recordings": 574,
   "reference_ap": 0.9702965395681998,
   "reference_model": "ced-base"
  }
 },
 "tolerance": {
  "ap": 1e-12,
  "ci": 0.005
 },
 "verdict": "FAIL: internal",
 "z_whole_ci": {
  "AMI": 6.409667762358887,
  "PF": 0.8831735739787857,
  "VS": -2.0794011367900485,
  "internal": 0.6112194331068893
 },
 "curation": {
  "by": "release packaging",
  "date": "2026-09-05",
  "note": "Public bundle, curated from the internal eval bundle: the private internal surface's per-clip data is excluded (its digest and aggregate row are kept so the card's internal row stays auditable), run names are mapped to release naming (laughter-v0.1-seed0..14, order preserved against the member_ap arrays), the fifteen per-seed configs ship once as configs/laughter-v0.1.json (runs are seeds 0-14 of it), and operational fields are removed from that config. eval/podcastfillers-relabel.tsv records the 100 PodcastFillers non-laughter clips both model families scored above 0.9 and the listening verdict on each (97 contain audible laughter); the table above uses the published labels. 2026-09-09: the FSD50K NSV panel added (fsd50k-panel.clips.tsv, probs for the 15 seeds and ced-base); its column is defined under fsd50k_panel. 2026-09-10: the two no-teacher arms retrained on seeds 0-14 so every trained row shares the same seeds; card_table holds every cell."
 },
 "fsd50k_panel": {
  "label": "FSD50K NSV",
  "what": "Share of FSD50K non-laughter vocalisation clips (any of cough, sneeze, breathing, gasp, sigh, crying, burping; no Laughter label) scored as laughter, at the threshold where the model recovers 90% of the panel's Laughter clips. Each clip's score is the max over its 4 s windows (hop 4 s). Lower is better.",
  "n_laughter": 933,
  "n_nsv": 1136,
  "n_clips": 4269,
  "recall": 0.9,
  "family_mean": 0.1459507042253521,
  "family_sd": 0.013560746885563212,
  "member_nsv_fp": [
   0.14964788732394366,
   0.15316901408450703,
   0.1267605633802817,
   0.13556338028169015,
   0.1637323943661972,
   0.13468309859154928,
   0.13556338028169015,
   0.1619718309859155,
   0.12147887323943662,
   0.13908450704225353,
   0.16549295774647887,
   0.14964788732394366,
   0.15933098591549297,
   0.14788732394366197,
   0.14524647887323944
  ],
  "reference_model": "ced-base",
  "reference_nsv_fp": 0.03609154929577465,
  "others": {
   "eat-lora-no-teacher": {
    "nsv_fp": 0.1931338028169014,
    "sd": 0.01694497340594801,
    "n": 15
   },
   "eat-full-finetune-teacher": {
    "nsv_fp": 0.17951877934272303,
    "sd": 0.04972447014311274,
    "n": 15
   },
   "eat-full-finetune-no-teacher": {
    "nsv_fp": 0.26085680751173707,
    "sd": 0.030258187344898973,
    "n": 15
   },
   "eat": {
    "nsv_fp": 0.0801056338028169,
    "n": 1
   },
   "ced-tiny": {
    "nsv_fp": 0.07130281690140845,
    "n": 1
   },
   "beats": {
    "nsv_fp": 0.14524647887323944,
    "n": 1
   },
   "gillick": {
    "nsv_fp": 0.48327464788732394,
    "n": 1
   }
  },
  "reproduce": "s = probs/fsd50k-panel__<model>.npy aligned with fsd50k-panel.clips.tsv; thr = the 840th highest score among laughter==1 rows (ceil(0.9*933)); nsv_fp = mean(s[nsv==1] >= thr). Clips are FSD50K dev audio (train and val folds) that carry panel labels; none of it is training data."
 },
 "card_table": {
  "note": "Every cell of the card's results table. Trained rows: mean and sd over seeds 0-14. Off-the-shelf rows: value and 95% bootstrap half-width (2000 draws, seed 0) over recordings (speakers on VocalSound, clips on the FSD50K panel). Built by training.card_results.",
  "columns": [
   "annotation-eval-v5",
   "podcastfillers-eval",
   "ami-eval-v1",
   "public-vocalsound",
   "fsd50k-panel"
  ],
  "panel_recall": 0.9,
  "bootstrap": 2000,
  "n_laughter": 933,
  "n_nsv": 1136,
  "rows": {
   "**this release**: EAT + LoRA, teacher-trained": {
    "annotation-eval-v5": {
     "value": 0.8996943593326668,
     "spread": 0.0029931705939927142,
     "spread_kind": "seed sd",
     "n": 15
    },
    "podcastfillers-eval": {
     "value": 0.8333527146573345,
     "spread": 0.010610126355847991,
     "spread_kind": "seed sd",
     "n": 15
    },
    "ami-eval-v1": {
     "value": 0.9483739882117678,
     "spread": 0.002510850524772026,
     "spread_kind": "seed sd",
     "n": 15
    },
    "public-vocalsound": {
     "value": 0.9550608658482184,
     "spread": 0.0046458419014504355,
     "spread_kind": "seed sd",
     "n": 15
    },
    "fsd50k-panel": {
     "value": 0.14595070422535214,
     "spread": 0.013560746885563212,
     "spread_kind": "seed sd",
     "n": 15
    }
   },
   "EAT + LoRA, no teacher": {
    "annotation-eval-v5": {
     "value": 0.9022181395955502,
     "spread": 0.0028596226591157932,
     "spread_kind": "seed sd",
     "n": 15
    },
    "podcastfillers-eval": {
     "value": 0.846728733967997,
     "spread": 0.014180319446455675,
     "spread_kind": "seed sd",
     "n": 15
    },
    "ami-eval-v1": {
     "value": 0.9524466643323456,
     "spread": 0.003276968221102732,
     "spread_kind": "seed sd",
     "n": 15
    },
    "public-vocalsound": {
     "value": 0.9420809875492873,
     "spread": 0.006998231467778743,
     "spread_kind": "seed sd",
     "n": 15
    },
    "fsd50k-panel": {
     "value": 0.1931338028169014,
     "spread": 0.01694497340594801,
     "spread_kind": "seed sd",
     "n": 15
    }
   },
   "EAT full fine-tune, teacher-trained": {
    "annotation-eval-v5": {
     "value": 0.9166429554840164,
     "spread": 0.005625359041947865,
     "spread_kind": "seed sd",
     "n": 15
    },
    "podcastfillers-eval": {
     "value": 0.8750703685982714,
     "spread": 0.011921685111299707,
     "spread_kind": "seed sd",
     "n": 15
    },
    "ami-eval-v1": {
     "value": 0.9446209554473263,
     "spread": 0.005212745805492586,
     "spread_kind": "seed sd",
     "n": 15
    },
    "public-vocalsound": {
     "value": 0.9437988023703398,
     "spread": 0.01700459450721973,
     "spread_kind": "seed sd",
     "n": 15
    },
    "fsd50k-panel": {
     "value": 0.17951877934272303,
     "spread": 0.04972447014311274,
     "spread_kind": "seed sd",
     "n": 15
    }
   },
   "EAT full fine-tune, no teacher": {
    "annotation-eval-v5": {
     "value": 0.918610648846479,
     "spread": 0.006374031205233863,
     "spread_kind": "seed sd",
     "n": 15
    },
    "podcastfillers-eval": {
     "value": 0.8808029276378652,
     "spread": 0.008482409285809667,
     "spread_kind": "seed sd",
     "n": 15
    },
    "ami-eval-v1": {
     "value": 0.9523479177219377,
     "spread": 0.004569827830633127,
     "spread_kind": "seed sd",
     "n": 15
    },
    "public-vocalsound": {
     "value": 0.9089073852987579,
     "spread": 0.059372105425155815,
     "spread_kind": "seed sd",
     "n": 15
    },
    "fsd50k-panel": {
     "value": 0.26085680751173707,
     "spread": 0.030258187344898973,
     "spread_kind": "seed sd",
     "n": 15
    }
   },
   "CED-base, AudioSet head": {
    "annotation-eval-v5": {
     "value": 0.8968090818701193,
     "spread": 0.0228192185597908,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "podcastfillers-eval": {
     "value": 0.8105831097976058,
     "spread": 0.06036189372814932,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "ami-eval-v1": {
     "value": 0.8541544974880243,
     "spread": 0.039408864655789466,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "public-vocalsound": {
     "value": 0.9702965395681998,
     "spread": 0.009631731710918856,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "fsd50k-panel": {
     "value": 0.03609154929577465,
     "spread": 0.016285211267605633,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    }
   },
   "EAT, AudioSet head (the base checkpoint, zero-shot)": {
    "annotation-eval-v5": {
     "value": 0.8829045193537908,
     "spread": 0.02558446961170535,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "podcastfillers-eval": {
     "value": 0.7844658804345496,
     "spread": 0.04947444610463847,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "ami-eval-v1": {
     "value": 0.9229926019779926,
     "spread": 0.024309727861856867,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "public-vocalsound": {
     "value": 0.9571953064111093,
     "spread": 0.013309956523687383,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "fsd50k-panel": {
     "value": 0.0801056338028169,
     "spread": 0.027739876760563317,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    }
   },
   "CED-tiny, AudioSet head": {
    "annotation-eval-v5": {
     "value": 0.8857709447136373,
     "spread": 0.02259637836585593,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "podcastfillers-eval": {
     "value": 0.7980219459500315,
     "spread": 0.06555965664476748,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "ami-eval-v1": {
     "value": 0.8564672033882755,
     "spread": 0.03207273515356468,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "public-vocalsound": {
     "value": 0.9558670210236383,
     "spread": 0.012283620217255986,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "fsd50k-panel": {
     "value": 0.07130281690140845,
     "spread": 0.02112676056338028,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    }
   },
   "BEATs, AudioSet head": {
    "annotation-eval-v5": {
     "value": 0.8383485927709583,
     "spread": 0.031978569016706926,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "podcastfillers-eval": {
     "value": 0.7475556060918509,
     "spread": 0.06449430520655708,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "ami-eval-v1": {
     "value": 0.8419949943494651,
     "spread": 0.029730828291308575,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "public-vocalsound": {
     "value": 0.9273063305843975,
     "spread": 0.016602341784197117,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "fsd50k-panel": {
     "value": 0.14524647887323944,
     "spread": 0.03962367957746479,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    }
   },
   "Gillick et al.": {
    "annotation-eval-v5": {
     "value": 0.8590035695842937,
     "spread": 0.022847874353034203,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "podcastfillers-eval": {
     "value": 0.8185743771254793,
     "spread": 0.04195457050713697,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "ami-eval-v1": {
     "value": 0.8525958378855036,
     "spread": 0.030944159938566085,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "public-vocalsound": {
     "value": 0.7171790828625292,
     "spread": 0.0312806324260928,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    },
    "fsd50k-panel": {
     "value": 0.48327464788732394,
     "spread": 0.053268045774647804,
     "spread_kind": "bootstrap 95% half-width",
     "n": 1
    }
   }
  }
 }
}