Plan Pointers and Record-Directive Form in Budgeted Verification of Inherited Agent Memory

Abstract. An agent that inherits six one-line memories may pull at most one archived source record before acting; a directive written into the store can steer that choice: a pointer to the record, a criterion that identifies it, or both. Across twelve registered studies on one instrument lineage (14,760 attempts) we measured where the request goes under each form. On six direct-provider models a length-matched criterion exceeded a bare id by +35.0 points +31.2, +38.8 (Study D); the contrast failed its registered superiority rule on a nine-model OpenRouter-served panel (Study E). Appending the id cancelled the criterion on three Claude models (Opus 5: 40/40 to 0/40; Study F-x); six byte-matched edits gave each exact string its own effect (Study G), and a re-run at eighty runs per cell left fifteen of thirty replication contrasts within the margin, fifteen unresolved and none beyond (Study G'). A ratification line (+96.0 points on Opus 5) and a budget of two credits restored the target on all three (Study J); across five criterion strings the suffix's cancellation held for four of the five wordings on Opus 5 and all five wordings on Fable 5.1 (Study H2); in a second store every model followed the criterion (Study H1). Continued into a decision, the criterion moved the choice toward the current record (+100.0 points, Opus 5) and away from it on Fable 5.1 (Study I). A one-character plan pointer's effect (+78.0 points; Study B, after a correction of its first repository report) returned the same verdict under a prospectively registered re-run (+81.7 points; Study B'). All results are descriptive effects of exact edits on fixed panels with registered intervals and no mechanism claim. Version notes (v1). Preprint, not peer reviewed. This record holds the manuscript (46 pages), its LaTeX source (identical to the arXiv source package) and the complete data-and-code archive of the twelve registered studies it reports (14,760 attempted episodes): every raw episode file with its completion manifest, every frozen registration package with its SHA-256 manifest, OpenTimestamps proof (upgraded to a Bitcoin block attestation) and OSF deposit receipt, the frozen analyzers and their registered outputs, the results reports and the post-freeze correction, deviation and erratum records, the manuscript generator that emits every number in the paper, the external review archives with their dispositions, and the literature search ledgers. README.md inside the archive explains the layout and how to verify and regenerate everything. Registration. Each study's package was hashed, committed, timestamped with OpenTimestamps and deposited to the OSF project axsnm before its first confirmatory call (the receipts carry the OSF server's creation time); each run was locked into a completion manifest before its single analysis. The six initial studies were outcome-sequential; the six follow-ups were conceived together and run one at a time. The OSF deposits are archival deposits made before execution, not registrations on a registry. Integrity. Study B's headline is a post-freeze correction on the locked episodes (the first repository report inverted the verdict by pooling a counterbalanced factor); Study F's first confirmatory run was contaminated by a provider-side web-search plugin and is quarantined in full inside the archive; every other deviation and erratum is recorded in the studies' files and summarised in the paper's integrity section. Related records. Paper 1: Verification Allocation in Inherited Agent Memory: Provenance Availability Is Not Provenance Use (concept DOI 10.5281/zenodo.22084498). Paper 2: When Stale Constraints Go Unchecked: Budgeted Verification Failures in Inherited Agent Memory (concept DOI 10.5281/zenodo.22108557; arXiv:2608.25553). OSF project holding the frozen packages: https://osf.io/axsnm/. AI assistance. Language-model assistants (Anthropic's Claude through Claude Code; OpenAI's ChatGPT/Codex) were used for research-design critique, implementation and execution of the runners, analysis and audit tooling, drafting and editing, and simulated adversarial review. The author chose the research questions, approved every experimental package and decided whether each run took place, interpreted the results, selected the claims, and is responsible for the correctness of the manuscript. No language model is an author; the models studied are experimental subjects, their responses are the data, and every outcome is scored deterministically. Licence. Manuscript, data and records: CC BY 4.0. Code: MIT (LICENSE inside the archive). Suggested citation. Nakayashiki, K. (2026). Plan Pointers and Record-Directive Form in Budgeted Verification of Inherited Agent Memory (v1). Zenodo. https://doi.org/10.5281/zenodo.22267221