{
  "format": "ta-amps.seed-pack.proposal.v1",
  "status": "proposal_only",
  "pack_id": "self-rewarding-and-iterative-dpo-flywheel",
  "title": "Self-rewarding and iterative DPO flywheel",
  "version": "0.1.0",
  "summary": "A collection of nouns and verbs for iterative RPO preference curation, rubric-judge calibration gating against deterministic verifiers and pre-registered holdout flywheel promotion.",
  "license": "MIT",
  "store": "seedpack.self-rewarding-and-iterative-dpo-flywheel",
  "queue_placeholder": "conformance-seed-packs",
  "files": [
    {
      "path": "README.md",
      "sha256": "d1e5df94abbd2f8fd2c20164cea47b428cfeefac17f880afda66483c149d1943"
    },
    {
      "path": "source.json",
      "sha256": "3a14664d206b543babe4705ea76ed7fded1f997384669f7d0977e43e278cf7fb"
    },
    {
      "path": "nouns.json",
      "sha256": "5a5e1beb02e0916d9ea857f6e6a55e8cf849dbdd694aa2341be78590e86cc5c5"
    },
    {
      "path": "verbs.json",
      "sha256": "17304759ae303c948c2139623c71c6e29b3e51bcbd319e842045721856798b6f"
    }
  ],
  "nouns": [
    {
      "key": "self-rewarding-and-iterative-dpo-flywheel.v0.1.0.pack",
      "category_id": "noun.skill_pack",
      "role": "pack",
      "value_sha256": "028d5a19b38952e46823245dff4e17e44a5898f46e40cef20f6e3df4992dee07"
    },
    {
      "key": "self-rewarding-and-iterative-dpo-flywheel.v0.1.0.guardrail.block-uncalibrated-synthetic-rubric-pairs",
      "category_id": "noun.skill_pack",
      "role": "guardrail",
      "value_sha256": "7cb46a0edcc6becf59dc750a792ab8e46ba519b613cf9ee2015c7b54e24df548"
    },
    {
      "key": "self-rewarding-and-iterative-dpo-flywheel.v0.1.0.guardrail.preregistered-holdout-lock-and-zero-magic-literals",
      "category_id": "noun.skill_pack",
      "role": "guardrail",
      "value_sha256": "9a8a38847ea3240ef297623357b91de674113fd803e2fc9c59165da601011cc7"
    },
    {
      "key": "self-rewarding-and-iterative-dpo-flywheel.v0.1.0.spec.iterative-rpo-and-flywheel-grid",
      "category_id": "noun.benchmark_suite",
      "role": "spec",
      "value_sha256": "7718341ff5d936823077ad1990255647603f58735bc00d10503a9c4606b14460"
    },
    {
      "key": "self-rewarding-and-iterative-dpo-flywheel.v0.1.0.rules.rpo-objective-and-calibration-gates",
      "category_id": "noun.reflex_procedure",
      "role": "rules",
      "value_sha256": "22235e9bb2d7131cafbcd0b9ecb917cb86cb9104d0d3386c359ff45539c90286"
    },
    {
      "key": "self-rewarding-and-iterative-dpo-flywheel.v0.1.0.tool.iterative-rpo-flywheel",
      "category_id": "noun.code_node",
      "role": "tool_card",
      "value_sha256": "4be9841095679303522ff8e2f28440424fe39b339a5fdf7461a5beb724042130"
    },
    {
      "key": "self-rewarding-and-iterative-dpo-flywheel.v0.1.0.template.iterative-rpo-flywheel-receipt",
      "category_id": "noun.scene_recipe",
      "role": "template",
      "value_sha256": "a1f71af1350671ca8cb681c98d855404a3bdcf5bb5cec750dbabb885c621d280"
    },
    {
      "key": "self-rewarding-and-iterative-dpo-flywheel.v0.1.0.example.iterative-rpo-flywheel-bundle",
      "category_id": "noun.eval_run",
      "role": "example",
      "value_sha256": "cbb7695f17b0eca8392637ab7f471a8ea1c906603e1f12e0566fb20c3bb7010d"
    },
    {
      "key": "self-rewarding-and-iterative-dpo-flywheel.v0.1.0.example.iterative-rpo-flywheel-receipt",
      "category_id": "noun.eval_run",
      "role": "example",
      "value_sha256": "41db6f9e916d7184fe5bd693b82b303f2837413819e1efc8623d4e3a8dbefa4c"
    }
  ],
  "verbs": [
    {
      "template_id": "self-rewarding-and-iterative-dpo-flywheel.calibrate-rubric-judge-against-verifiers",
      "kind": "verb.probe_source",
      "claimant": "agent",
      "idempotency_key": "seedpack:self-rewarding-and-iterative-dpo-flywheel:calibrate-rubric-judge-against-verifiers:bebb5071d735a0e617e569f6551c4dfd515faa817990f8dc0219db91b5c315ea",
      "report_shape_sha256": "ee3da5c4d491d860ecd5b06f2dafe158f050384f3af9e50e1a9abfba77819e1a"
    },
    {
      "template_id": "self-rewarding-and-iterative-dpo-flywheel.curate-iterative-rpo-preference-pairs",
      "kind": "verb.generate_seed",
      "claimant": "agent",
      "idempotency_key": "seedpack:self-rewarding-and-iterative-dpo-flywheel:curate-iterative-rpo-preference-pairs:f8db7286eb8c0b39898edb7f95d17ef25515c38110bc2268d6384d1151c36afb",
      "report_shape_sha256": "f02bf4d8a98469e9e5d71cbb649619800c8b355d7f23a53ff4f77939a71fc3f0"
    },
    {
      "template_id": "self-rewarding-and-iterative-dpo-flywheel.verify-preregistered-flywheel-promotion",
      "kind": "verb.verify_gate",
      "claimant": "agent",
      "idempotency_key": "seedpack:self-rewarding-and-iterative-dpo-flywheel:verify-preregistered-flywheel-promotion:ff1070d666b8de31a19587deb2de5a51640a53414eab78f8f21e0ad87a7cefcd",
      "report_shape_sha256": "239eada750ba0f5f8507722c2b97cd72c9e7eeb4893e41c18fe6d2c46ef6d3ee"
    }
  ],
  "code": {
    "path": "packs/self-rewarding-and-iterative-dpo-flywheel/code",
    "files": 12,
    "sha256": "e1da62b88a09486fd09380cbd2f67238f9f4f5c204a2df85f7f29fffb5658275",
    "runs": "The library runs this code only in its own CI tests; workers run it under their own permissions. Reading, exporting or importing the pack runs none of it."
  },
  "dependencies": [
    {
      "pack_id": "configuration-grids-and-bounds",
      "keys": []
    },
    {
      "pack_id": "evaluation-design",
      "keys": []
    },
    {
      "pack_id": "wiring-and-conformance-checks",
      "keys": []
    }
  ],
  "boundaries": [
    "RubricJudgeCalibrationGate blocks synthetic self-rewarding preference pairs whenever Cohen's kappa, exact agreement or false-positive approval rates violate grid calibration bounds against deterministic verifiers.",
    "IterativeRpoPairCurator enforces positive NLL_chosen regularization, length-ratio bias bounds and 3-key structural deduplication so iterative preference rounds do not collapse chosen log-probabilities or reward verbosity.",
    "PreRegisteredFlywheelPromoter rejects candidate generation rounds upon any holdout SHA-256 mismatch, self-review, sanity retention drop or regression cap breach.",
    "Importing the pack writes nothing, enqueues nothing and installs no code."
  ],
  "override_policy": {
    "automatic_merge": false,
    "on_conflict": "stop_and_clarify"
  },
  "importer": null,
  "importer_reason": "No manifest endpoint or generic importer exists. Reading or exporting this pack writes no records, enqueues no verbs and activates nothing.",
  "activation": {
    "automatic": false,
    "steps": [
      "An owner assigns the namespace, store and queue.",
      "Each noun is written create-only with existing grants and read back; its value SHA-256 must match the pin.",
      "A verb template is filled with exact pins and enqueued explicitly by an authorized requester."
    ]
  },
  "evaluation": {
    "result": null,
    "reason": "Offline shape, category and digest validation only. No worker has run these verbs."
  }
}