# kaal:position:2026-08-08-357

**Affirmed position.** 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.

The 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.

**Status.** affirmed  **Published.** 2026-08-08

**Holds when.**

- 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.

**Current debate.** Population Based Training of Neural Networks: https://arxiv.org/abs/1711.09846

**Extends.** kaal:claim:7261481-022: https://wulfkaal.github.io/claims/7261481-022

**Scholarly basis.** Wulf A. Kaal, Computative Economics: A Framework for Economic Analysis under Computational Abundance (2026). SSRN: https://ssrn.com/abstract=7261481

**Source PDF sha256.** `78c42db521624f7398717732a7fa51a6e3157a5adf02a2e09fbab15e0cf920d9`

**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

**Mapping confidence.** 0.97  **Mapping ambiguous.** false

**Topics.** research-methods, ai-and-agents, scholarly-growth-coverage, scholarly-literature, population-based-training, evolutionary-agents, lineage, selection-rules

**Provenance.** Affirmed in kaal-review:2026-08-13:scholarly-growth-7261481-022-reviewed-v1 at https://wulfkaal.github.io/positions/by-claim/7261481-022.html.

**Record type.** This is a dated commentary position that extends a scholarly corpus claim. It is not a verbatim claim extracted from the paper.

**Canonical form.** This markdown file is the canonical hashed representation of the position.