{
 "model": "D1A-E2B v0.6",
 "base": "google/gemma-4-E2B",
 "base_revision": "d29ff6b45f081a49ee2733a859c9c9c2d95d1a6f",
 "init_from": "JohnP1/d1a-e2b@v0.2.1-2epoch-calibrated (commit b529d2fd25bb813319903ff8f1304e670677a6bb; v0.2.1 had only decision-v7)",
 "stage": "Every later skill in one run (jonpol01/d1a#229): PR labels (English 4,000, Japanese 753, blast radius 1,500, and 479 hermes-agent PRs created 2026-10-02 to 10-05 shown twice), hard-v1 4,000, devtools-v1 4,000, documents-v1 2,500, JGLUE 2,000, routing factory/generic 1,216/1,000, decision-v7 1,500; 23,427 records; 1 epoch, lr 5e-5, LoRA 16 on all projections, batch 2 x accum 4, states up to 5,120 tokens, 2,929 steps",
 "release_exception": "Ships despite #229's pre-registered bar (no card suite below -1: decision-v7 -1.4 [-2.8, -0.1]), by the owner's decision of 2026-10-11, so that E2B and E4B each exist as MLX and Core ML builds and can be compared. E2B v0.6 is not a replacement for E4B v0.6 as the PR labeler (P0/P1 recall 0/30 on newer PRs).",
 "data": {
  "counts": {
   "pr-labels:fresh-fit": {
    "records": 958,
    "pool_after_guards": 479,
    "fill": null,
    "labels": {
     "type": {
      "type/bug": 310,
      "type/feature": 112,
      "type/security": 18,
      "type/refactor": 16,
      "type/test": 11,
      "type/docs": 11,
      "type/perf": 1
     },
     "sev": {
      "P3": 257,
      "P2": 132,
      "P1": 65,
      "P4": 13,
      "P0": 12
     }
    },
    "reuse": {
     "never trained": 479
    },
    "est_tokens": 522078,
    "est_tok_per_rec": 1090,
    "est_mps_hours": 2.05,
    "passes": 2
   },
   "pr-labels:train": {
    "records": 4000,
    "pool_after_guards": 7386,
    "fill": {
     "P3": 2243,
     "P2": 924,
     "P4": 457,
     "P1": 208,
     "P0": 165,
     "None": 3
    },
    "labels": {
     "type": {
      "type/bug": 797,
      "type/feature": 607,
      "type/security": 602,
      "type/docs": 588,
      "type/perf": 563,
      "type/test": 555,
      "type/refactor": 288
     },
     "sev": {
      "P3": 2243,
      "P2": 924,
      "P4": 457,
      "P1": 208,
      "P0": 165
     },
     "blast": {
      "review:blast-moderate": 208,
      "review:blast-contained": 193,
      "review:blast-broad": 189,
      "review:blast-massive": 33
     }
    },
    "reuse": {
     "v0.4 (pr-labeler round 2)": 3419,
     "v0.3 (pr-labeler round 1)": 1929,
     "v0.5 (skills stage)": 721,
     "phase C (running)": 473,
     "B0 pilot (v0.4 + 300 steps)": 175
    },
    "est_tokens": 3818332,
    "est_tok_per_rec": 955,
    "est_mps_hours": 15.0
   },
   "pr-labels:train-ja": {
    "records": 753,
    "pool_after_guards": 753,
    "fill": null,
    "labels": {
     "type": {
      "type/bug": 149,
      "type/feature": 121,
      "type/perf": 104,
      "type/security": 98,
      "type/refactor": 97,
      "type/docs": 95,
      "type/test": 89
     },
     "sev": {
      "P3": 450,
      "P2": 162,
      "P4": 77,
      "P1": 31,
      "P0": 31
     },
     "blast": {
      "review:blast-broad": 41,
      "review:blast-moderate": 29,
      "review:blast-contained": 28,
      "review:blast-massive": 6
     }
    },
    "reuse": {
     "v0.3 (pr-labeler round 1)": 753,
     "v0.4 (pr-labeler round 2)": 753,
     "phase C (running)": 245
    },
    "est_tokens": 468621,
    "est_tok_per_rec": 622,
    "est_mps_hours": 1.84
   },
   "pr-labels:train-blast": {
    "records": 1500,
    "pool_after_guards": 2185,
    "fill": {
     "review:blast-contained": 630,
     "review:blast-moderate": 608,
     "review:blast-broad": 262
    },
    "labels": {
     "blast": {
      "review:blast-contained": 630,
      "review:blast-moderate": 608,
      "review:blast-broad": 262
     },
     "sev": {
      "P3": 860,
      "P2": 486,
      "P1": 143,
      "P4": 10
     }
    },
    "reuse": {
     "v0.4 (pr-labeler round 2)": 1500,
     "phase C (running)": 1238
    },
    "est_tokens": 1304009,
    "est_tok_per_rec": 869,
    "est_mps_hours": 5.12
   },
   "hard-v1:train": {
    "records": 4000,
    "pool_after_guards": 5996,
    "fill": {
     "tradeoff": 571,
     "temporal_numeric": 569,
     "ambiguous": 572,
     "multi_hop": 572,
     "probability": 572,
     "judge": 572,
     "long_policy": 572
    },
    "labels": {
     "_meta.family": {
      "ambiguous": 572,
      "judge": 572,
      "long_policy": 572,
      "multi_hop": 572,
      "probability": 572,
      "tradeoff": 571,
      "temporal_numeric": 569
     },
     "_meta.source": {
      "hard_ambiguous": 572,
      "hard_judge": 572,
      "hard_long_policy": 572,
      "hard_multi_hop": 572,
      "hard_probability": 572,
      "hard_tradeoff": 571,
      "hard_temporal_numeric": 569
     }
    },
    "reuse": {
     "v0.5 (skills stage)": 4000
    },
    "est_tokens": 2842243,
    "est_tok_per_rec": 711,
    "est_mps_hours": 11.17
   },
   "devtools-v1:train": {
    "records": 4000,
    "pool_after_guards": 5314,
    "fill": {
     "flakeflagger": 617,
     "commitpackft": 1130,
     "codereviewer": 1128,
     "aegis": 1125
    },
    "labels": {
     "_meta.source": {
      "commitpackft": 1130,
      "codereviewer": 1128,
      "aegis": 1125,
      "flakeflagger": 617
     }
    },
    "reuse": {
     "v0.5 (skills stage)": 4000
    },
    "est_tokens": 1080361,
    "est_tok_per_rec": 270,
    "est_mps_hours": 4.24
   },
   "documents-v1:train": {
    "records": 2500,
    "pool_after_guards": 5120,
    "fill": {
     "debt_collection": 167,
     "card": 275,
     "vehicle_loan": 280,
     "money_transfer": 209,
     "credit_reporting": 252,
     "mortgage": 314,
     "bank_account": 251,
     "None": 243,
     "student_loan": 300,
     "personal_loan": 209
    },
    "labels": {
     "issue": {
      "lender_or_servicer": 213,
      "managing": 144,
      "payment_process": 107,
      "fraud_or_scam": 93,
      "incorrect_information": 86,
      "debt_not_owed": 78,
      "purchase_on_statement": 78,
      "closing": 54,
      "getting_a_loan": 46,
      "making_payments": 42,
      "struggling_to_pay": 39,
      "unexpected_fees": 34,
      "end_of_loan": 31,
      "charged_by_another_company": 28,
      "money_not_available": 23,
      "closing_the_account": 23,
      "communication_tactics": 22,
      "getting_the_loan": 22,
      "applying": 21,
      "unauthorized_transactions": 20,
      "payoff": 20,
      "improper_use": 20,
      "investigation": 20,
      "fees_or_interest": 18,
      "struggling_to_repay": 15,
      "getting_a_card": 12,
      "repossession": 12,
      "opening": 10,
      "written_notification": 8,
      "mobile_wallet_account": 6,
      "false_statements": 3,
      "negative_or_legal_action": 2,
      "fraud_alerts_or_freezes": 2,
      "low_funds": 1,
      "unable_to_get_report": 1
     },
     "_meta.source": {
      "cfpb": 2500
     },
     "product": {
      "mortgage": 314,
      "student_loan": 300,
      "vehicle_loan": 280,
      "card": 275,
      "credit_reporting": 252,
      "bank_account": 251,
      "money_transfer": 209,
      "personal_loan": 209,
      "debt_collection": 167
     }
    },
    "reuse": {
     "never trained": 2500
    },
    "est_tokens": 2357246,
    "est_tok_per_rec": 943,
    "est_mps_hours": 9.26
   },
   "ja-jglue:train": {
    "records": 2000,
    "pool_after_guards": 5983,
    "fill": null,
    "labels": {},
    "reuse": {
     "v0.2 run 2 (JGLUE all + routing all + dv7 replay 3000)": 2000
    },
    "est_tokens": 118097,
    "est_tok_per_rec": 59,
    "est_mps_hours": 0.46
   },
   "routing:factory-train": {
    "records": 1216,
    "pool_after_guards": 1216,
    "fill": null,
    "labels": {
     "tier": {
      "medium": 234,
      "small": 204,
      "large": 174
     },
     "decision": {
      "allow": 123,
      "ask": 99,
      "deny": 84
     },
     "worker": {
      "developer": 149,
      "handler": 46,
      "mechanic": 39,
      "avatar": 33,
      "broker": 22,
      "recruiter": 17
     },
     "has_target": {
      "True": 215,
      "False": 91
     },
     "has_spec": {
      "True": 185,
      "False": 121
     },
     "clear_done": {
      "True": 185,
      "False": 121
     },
     "consult": {
      "False": 183,
      "True": 123
     },
     "done": {
      "False": 214,
      "True": 84
     },
     "blocked": {
      "False": 176,
      "True": 122
     },
     "human": {
      "False": 217,
      "True": 81
     },
     "next": {
      "rework": 92,
      "complete": 84,
      "consult": 81,
      "repair": 41
     }
    },
    "reuse": {
     "v0.2 run 2 (JGLUE all + routing all + dv7 replay 3000)": 1216,
     "v0.3 (pr-labeler round 1)": 141,
     "v0.4 (pr-labeler round 2)": 141,
     "phase C (running)": 125
    },
    "est_tokens": 328029,
    "est_tok_per_rec": 270,
    "est_mps_hours": 1.29
   },
   "routing:generic-train": {
    "records": 1000,
    "pool_after_guards": 1529,
    "fill": null,
    "labels": {
     "route": {
      "small": 339,
      "medium": 338,
      "large": 323
     }
    },
    "reuse": {
     "v0.2 run 2 (JGLUE all + routing all + dv7 replay 3000)": 1000
    },
    "est_tokens": 142874,
    "est_tok_per_rec": 143,
    "est_mps_hours": 0.56
   },
   "decision-v7:train": {
    "records": 1500,
    "pool_after_guards": 11219,
    "fill": {
     "stars": 132,
     "intent": 112,
     "answer": 132,
     "answer_type": 114,
     "positive": 115,
     "category": 115,
     "rating+recommend": 115,
     "relation": 132,
     "decision": 283,
     "is_business+is_world+topic": 18,
     "is_sports+is_world+topic": 20,
     "sentiment": 133,
     "is_scitech+is_sports+topic": 20,
     "is_business+is_scitech+topic": 21,
     "is_scitech+is_world+topic": 20,
     "is_business+is_sports+topic": 18
    },
    "labels": {
     "_meta.source": {
      "legacy_policy": 142,
      "compositional": 141,
      "sst5": 133,
      "boolq": 132,
      "mnli": 132,
      "amazon": 132,
      "agnews": 117,
      "dbpedia14": 115,
      "imdb": 115,
      "yelp": 115,
      "trec": 114,
      "banking77": 112
     },
     "_meta.family": {
      "late_fee": 25,
      "sla_response": 23,
      "spend_threshold": 19,
      "quantity_limit": 18,
      "return_window": 18,
      "warranty_claim": 15,
      "volume_discount": 15,
      "age_eligibility": 9,
      "rand:unless(and(0,1),and(2,3))": 6,
      "rand:if(0,and(1,2),not(3))": 6,
      "rand:and(and(0,1),and(2,3))": 4,
      "rand:or(and(0,1),unless(2,3))": 4,
      "rand:or(0,or(and(1,2),not(3)))": 4,
      "rand:if(and(0,1),2,3)": 4,
      "rand:unless(or(not(0),unless(1,2)),3)": 4,
      "rand:unless(or(0,1),2)": 4,
      "rand:or(0,or(1,or(2,3)))": 4,
      "rand:if(0,not(1),and(2,3))": 4,
      "rand:and(or(0,1),or(2,3))": 3,
      "rand:or(not(0),unless(1,2))": 3,
      "rand:or(0,not(or(1,2)))": 3,
      "rand:if(0,not(or(1,2)),not(3))": 3,
      "rand:and(0,unless(1,2))": 3,
      "rand:or(0,if(1,2,3))": 3,
      "rand:not(if(0,not(1),or(2,3)))": 3,
      "rand:unless(0,and(1,2))": 3,
      "rand:if(0,and(not(1),not(2)),not(not(3)))": 3,
      "rand:and(0,not(if(1,2,3)))": 3,
      "rand:unless(and(0,1),2)": 3,
      "rand:or(or(0,1),unless(2,3))": 3,
      "rand:unless(0,or(1,2))": 3,
      "rand:and(0,1)": 3,
      "rand:unless(not(or(0,1)),not(and(2,3)))": 3,
      "rand:or(and(0,1),and(2,3))": 3,
      "rand:unless(0,and(and(1,2),not(3)))": 2,
      "rand:and(0,and(1,or(2,3)))": 2,
      "rand:and(and(0,not(1)),and(2,not(3)))": 2,
      "rand:not(unless(0,1))": 2,
      "rand:unless(0,and(1,and(2,3)))": 2,
      "rand:unless(unless(0,1),2)": 2,
      "rand:not(or(0,or(1,2)))": 2,
      "rand:if(0,1,and(2,3))": 2,
      "rand:or(or(0,1),or(2,3))": 2,
      "rand:unless(or(0,1),or(2,3))": 2,
      "rand:unless(or(0,1),and(2,3))": 2,
      "rand:or(0,and(1,2))": 2,
      "rand:or(unless(0,1),unless(2,3))": 2,
      "rand:if(not(0),1,2)": 2,
      "rand:not(not(or(0,1)))": 2,
      "rand:and(0,unless(1,not(2)))": 2,
      "rand:or(if(0,1,2),not(3))": 2,
      "rand:unless(and(0,1),unless(2,3))": 2,
      "rand:or(0,or(1,unless(2,3)))": 2,
      "rand:and(0,not(1))": 2,
      "rand:not(if(0,1,2))": 1,
      "rand:and(and(0,1),or(2,3))": 1,
      "rand:not(and(0,or(1,2)))": 1,
      "rand:or(0,or(1,2))": 1,
      "rand:or(0,not(and(1,2)))": 1,
      "rand:and(if(0,1,2),not(3))": 1,
      "rand:not(and(0,1))": 1,
      "rand:unless(unless(0,or(1,2)),3)": 1,
      "rand:and(and(0,1),unless(2,3))": 1
     }
    },
    "reuse": {
     "v0.1 (decision-v7 suite, 2 epochs)": 1500
    },
    "est_tokens": 358990,
    "est_tok_per_rec": 239,
    "est_mps_hours": 1.41
   }
  },
  "records": 23427,
  "sha256": "efff4b161e2d493f3d6e33712174614964279446ed2deb65b78ad4f7cb86e7a9",
  "seed": 0,
  "fresh_prs": {
   "prs": 479,
   "passes": 2,
   "created": "2026-10-02 to 2026-10-05",
   "text": "private; aggregates only"
  },
  "leak_checks": "0 eval-state overlap, 0 PR-id overlap with the sealed fresh EVAL half and every eval partition"
 },
 "training": {
  "epochs": 1,
  "lr": 5e-05,
  "head_lr": 0.0,
  "weight_decay": 0.01,
  "lora": 16,
  "lora_targets": "all",
  "batch": 2,
  "accum": 4,
  "dtype": "bf16",
  "weights_dtype": "bf16",
  "checkpointing": 1,
  "max_state": 5120,
  "p_none": 0.1,
  "p_none_distract": 0.12,
  "p_distract": 0.15,
  "seed": 0
 },
 "compute": {
  "where": "Hugging Face Jobs, one NVIDIA L40S (48 GB)",
  "wall_seconds": 6163,
  "optimizer_steps": 2929,
  "records_seen": 23427,
  "truncated_records": 0,
  "rejected_records": 0
 },
 "calibration": {
  "temperature": 1.447,
  "fit": "pooled rows: decision-v7 calibration split + PR-labeling development set",
  "interval_90": [
   1.382,
   1.516
  ],
  "ece_raw": 0.051,
  "ece_out_of_fold": 0.024,
  "use_case_temperatures": "none"
 },
 "code": "github.com/jonpol01/d1a (d1a.training.train, scripts/quality_gate.py, scripts/decide.py)"
}