{
 "@context": "https://schema.org",
 "@type": "Claim",
 "@id": "https://wulfkaal.github.io/positions/2026-08-08-357",
 "identifier": "kaal:position:2026-08-08-357",
 "additionalType": "https://wulfkaal.github.io/positions/schema.json#AffirmedPositionClaim",
 "name": "Population Based Training Qualifies Selection Induced Lineage Transition",
 "text": "Jaderberg, Dalibard, Osindero, Czarnecki, and their coauthors provide an independent operational comparator for a selection-induced lineage transition. The mechanism is concrete. Population Based Training evaluates concurrently trained agents and applies an exploit rule when a population member is ready. Under one reported rule, a lower-performing agent copies the weights and hyperparameters of a better-performing agent, while another rule directs an agent in the bottom fifth to copy from the top fifth. Training then continues from the inherited model state. The authors represent each executed exploit operation as a branch in the full phylogenetic tree because the parameters were copied. The transition is therefore an implemented change in model ancestry rather than a descriptive label.\n\nThe relationship is a qualification. The paper shows that a performance-based selection rule can produce a traceable computational lineage transition. It does not validate Kaal's registered E2B rule, fitness measure, parentage record, or execution receipts. Population Based Training also copies model weights and hyperparameters within an asynchronous optimizer. Kaal's lineage objects and selection conditions may differ. The external evidence establishes mechanism correspondence and operational feasibility. Historical occurrence in Kaal's experiment remains bound to the interim E2B record. A lineage transition is externally legible only when the selection event and inherited state are both preserved.",
 "author": {
  "@type": "Person",
  "name": "Wulf A. Kaal",
  "identifier": "https://orcid.org/0009-0008-7840-1847"
 },
 "datePublished": "2026-08-08",
 "dateModified": "2026-08-08",
 "creativeWorkStatus": "Affirmed",
 "responseType": "qualification",
 "keywords": [
  "research-methods",
  "ai-and-agents",
  "scholarly-growth-coverage",
  "scholarly-literature",
  "population-based-training",
  "evolutionary-agents",
  "lineage",
  "selection-rules"
 ],
 "scope_conditions": [
  "The response is limited to the exact public primary-paper proposition and the one mapped Kaal claim.",
  "External evidence level: complete public 21-page arXiv v2 primary paper with exact PDF SHA-256, page-bound passages, arXiv API identity, and DataCite identity.",
  "Mapping review tier: independent substantive scholarly-growth qualification.",
  "The source does not inspect or validate Kaal's registered E2B rule, fitness measure, parentage record, or execution receipts.",
  "Population Based Training copies weights and hyperparameters within an asynchronous optimizer, while Kaal's lineage objects and selection conditions may differ.",
  "The source establishes mechanism correspondence and operational feasibility, not historical occurrence in Kaal's experiment.",
  "The retained item is an arXiv v2 primary paper and is not represented here as a peer-reviewed journal article."
 ],
 "currentDebate": {
  "name": "Population Based Training of Neural Networks",
  "url": "https://arxiv.org/abs/1711.09846"
 },
 "extends": {
  "identifier": "kaal:claim:7261481-022",
  "url": "https://wulfkaal.github.io/claims/7261481-022",
  "citation": "Wulf A. Kaal, Computative Economics: A Framework for Economic Analysis under Computational Abundance (2026). SSRN: https://ssrn.com/abstract=7261481",
  "paper": "Wulf A. Kaal, Computative Economics: A Framework for Economic Analysis under Computational Abundance",
  "authors": [
   "Wulf A. Kaal"
  ],
  "year": "2026",
  "ssrn": "https://ssrn.com/abstract=7261481",
  "source_pdf_sha256": "78c42db521624f7398717732a7fa51a6e3157a5adf02a2e09fbab15e0cf920d9"
 },
 "isBasedOn": [
  {
   "@id": "https://wulfkaal.github.io/claims/7261481-022"
  },
  {
   "@type": "CreativeWork",
   "name": "Population Based Training of Neural Networks",
   "url": "https://arxiv.org/abs/1711.09846"
  }
 ],
 "batch_id": "kaal-review:2026-08-13:scholarly-growth-7261481-022-reviewed-v1",
 "review_provenance": "https://wulfkaal.github.io/positions/by-claim/7261481-022.html",
 "publicationStatus": "public",
 "recordTypeNote": "Dated commentary position extending a scholarly corpus claim. Not a verbatim claim extracted from the paper.",
 "isPartOf": {
  "@id": "https://wulfkaal.github.io/positions/index.json"
 },
 "version": "1.0",
 "canonical_url": "https://wulfkaal.github.io/positions/2026-08-08-357",
 "canonicalForm": "https://wulfkaal.github.io/positions/2026-08-08-357.md",
 "candidateId": "kaal:response-candidate:2026-08-13:scholarly-growth-7261481-022-population-based-training-01",
 "evidenceLevel": "complete public 21-page arXiv v2 primary paper with exact PDF SHA-256, page-bound passages, arXiv API identity, and DataCite identity",
 "reviewTier": "independent substantive scholarly-growth qualification",
 "mappingConfidence": 0.97,
 "mappingAmbiguous": false,
 "mappingMethod": "independent substantive scholarly-growth one-to-one review",
 "mappingWhyRelevant": "The source directly links performance-based selection, parameter inheritance, continued training, and a phylogenetic branch while leaving Kaal's registered rule and historical execution source-bound.",
 "sourceProvenance": {
  "source": "complete public arXiv v2 primary paper independently bound through PDF, arXiv abstract page, arXiv API, and DataCite",
  "sourceRecordId": "arxiv:1711.09846v2",
  "canonicalUrl": "https://doi.org/10.48550/arXiv.1711.09846",
  "publicFullTextUrl": "https://arxiv.org/pdf/1711.09846",
  "retrievedAt": "2026-08-13T17:39:49.280Z",
  "sourceArticlePages": 21,
  "primaryFullTextEvidenceReceiptSha256": "190ec14bc682790c7f77ac455120411534f55ff83770f9b46ee5c7cb23f6ae1a",
  "pdfSha256": "47db40f95ba20505e8557252bd53d6d9c9a98b8c34f8e92b6db30d59ca993e93",
  "extractedTextSha256": "22f6baff3f3ec2d93805777483664f9aa46261bb012cf8909689f58bf587a435",
  "rawExtractedTextSha256": "5f5a6710480ca065b5f9d46c02daa9a4f5a3c3ce8cf4a2b6424fad37e381c3ca",
  "arxivAbstractSha256": "5f8bf42b475216c59a19107edbbf0ac389b3e8e40fb394ac87a3b8203ac2e974",
  "arxivApiSha256": "21469f24a887ee7832c471bbff893cca3aedea0658756264097975ceba219ebd",
  "dataCiteResponseSha256": "727c28f9acd42ef6ea69f507d061e6c74f328196ec0cd1b826b86bac1e7522a8",
  "textExtraction": {
   "tool": "pdftotext 25.06.0 with layout and raw preservation",
   "quality": "complete readable 21-page primary paper with algorithm, experiments, phylogenetic analysis, appendices, and exact page-bound passages"
  },
  "sourceProposition": "Jaderberg and coauthors implement performance-based exploit rules that replace an underperforming agent's weights and hyperparameters with those of a better-performing agent. Their full phylogenetic development graph represents each executed copy as a branch, making the selection-induced model lineage transition observable.",
  "sourcePropositionSha256": "ad8429bc0f5e8fa631d34e9b7d5279abf3ce3951bc94e7f90e9cad279c4c806b",
  "sourceEvidenceSetSha256": "f6ce4c3b6d6c6871dad8962f7604242076cafc5bafcc7e23cfa8434e7c561fde",
  "sourceEvidencePassages": [
   {
    "text": "For example, exploit could replace the current weights with the weights that have the highest recorded performance in the rest of the population, and explore could randomly perturb the hyperparameters with noise.",
    "locator": {
     "version": "arXiv:1711.09846v2",
     "printedPage": 5,
     "section": "3 Population Based Training",
     "primaryFullTextUrl": "https://arxiv.org/pdf/1711.09846"
    },
    "sha256": "4ca88871901270488b0c1d9aa5645290ee3a79bcb18d14ceaff7d24a15b7bcdf"
   },
   {
    "text": "If the sampled agent has a higher mean episodic reward and satisfies the t-test, the weights and hyperparameters are copied to replace the current agent.",
    "locator": {
     "version": "arXiv:1711.09846v2",
     "printedPage": 7,
     "section": "4.1.1 PBT for RL",
     "primaryFullTextUrl": "https://arxiv.org/pdf/1711.09846"
    },
    "sha256": "cb48a9afc9d46d4164ba648b3aa2cec7bbb4f0acce0926e7d0692370852697d5"
   },
   {
    "text": "If the current agent is in the bottom 20% of the population, we sample another agent uniformly from the top 20% of the population, and copy its weights and hyperparameters.",
    "locator": {
     "version": "arXiv:1711.09846v2",
     "printedPage": 7,
     "section": "4.1.1 PBT for RL",
     "primaryFullTextUrl": "https://arxiv.org/pdf/1711.09846"
    },
    "sha256": "7613c24e52db8b3e59f80ff379ecc9175cbfcdaf0b8025f19cef538573e16e00"
   },
   {
    "text": "Branching of the graph means that the exploit operation has been executed (and so parameters were copied), while paths represent consecutive updates using the step function.",
    "locator": {
     "version": "arXiv:1711.09846v2",
     "printedPage": 9,
     "section": "Figure 6 caption",
     "primaryFullTextUrl": "https://arxiv.org/pdf/1711.09846"
    },
    "sha256": "954b5db464492796d5cac64e9aa1cc0a1493ca75d8a67e1bc020117ad3a3ff4c"
   }
  ],
  "workId": "work:arxiv:1711.09846v2",
  "workAuthors": [
   "Max Jaderberg",
   "Valentin Dalibard",
   "Simon Osindero",
   "Wojciech M. Czarnecki",
   "Jeff Donahue",
   "Ali Razavi",
   "Oriol Vinyals",
   "Tim Green",
   "Iain Dunning",
   "Karen Simonyan",
   "Chrisantha Fernando",
   "Koray Kavukcuoglu"
  ],
  "workPublishedAt": "2017-11-27",
  "identityKeys": [
   "arxiv:1711.09846v2",
   "doi:10.48550/arxiv.1711.09846",
   "pdf:47db40f95ba20505e8557252bd53d6d9c9a98b8c34f8e92b6db30d59ca993e93",
   "proposition:ad8429bc0f5e8fa631d34e9b7d5279abf3ce3951bc94e7f90e9cad279c4c806b"
  ],
  "claimMappings": [
   {
    "claimId": "kaal:claim:7261481-022",
    "claimUrl": "https://wulfkaal.github.io/claims/7261481-022",
    "rank": 1,
    "confidence": 0.97,
    "method": "independent substantive scholarly-growth one-to-one review",
    "whyRelevant": "The source directly links performance-based selection, parameter inheritance, continued training, and a phylogenetic branch while leaving Kaal's registered rule and historical execution source-bound.",
    "ambiguous": false
   }
  ],
  "substantiveReview": {
   "reviewedAt": "2026-08-13T17:44:58.656Z",
   "sourceIdentityVerified": true,
   "authorIndependenceVerified": true,
   "canonicalPublicStatusVerified": true,
   "retractionOrSupersessionFound": false,
   "propositionFidelityVerified": true,
   "mechanismCorrespondence": "performance-triggered selection copies a better agent's inherited weights and hyperparameters, continues training, and records the copy as a phylogenetic branch",
   "compatibleScope": "qualification limited to an external operational comparator for a selection-induced computational lineage transition",
   "responseWordingDefensible": true,
   "oneToOneExtendsMapping": true,
   "sameSourceDistinctionVerified": true,
   "limitations": [
    "The source does not inspect or validate Kaal's registered E2B rule, fitness measure, parentage record, or execution receipts.",
    "Population Based Training copies weights and hyperparameters within an asynchronous optimizer, while Kaal's lineage objects and selection conditions may differ.",
    "The source establishes mechanism correspondence and operational feasibility, not historical occurrence in Kaal's experiment.",
    "The retained item is an arXiv v2 primary paper and is not represented here as a peer-reviewed journal article."
   ],
   "exactSupportingQuotesVerified": true
  }
 },
 "userAffirmation": "Automatically authorized under standing authority receipt kaal-standing-publication-authorization:2026-08-01:hourly-reviewed-batches, SHA-256 e2126054b58bb4e88db65c334ef4fc8ae78dcaadc63543c408133c4eefceb9b1, limited to this substantively reviewed scholarly-growth qualification and the protected scholarly claims matching the current bridge checkpoint under the current owner instruction.",
 "sha256": "14b48c9b1fc7fdabbc8400308a3150b3c8cf5d7b043bc8b0aa7d53fcc7781915"
}