Do agent memories keep an exception's scope? Protocol and materials frozen before the run (Study 5)

Frozen protocol, code and materials of a study of whether the memories that LLM agents' memory-writing configurations write keep the scope of a temporary exception, deposited before any smoke or confirmatory model call (the manifest's OpenTimestamps proof is included; its Bitcoin attestation is checked by the gated launcher). On scripted histories of a recurring operational task, a temporary exception (a request to use a fallback route on stated dates, or an outage of the standard route) either still applies at the next scheduled run or has ended; the two versions of each history differ only in that. Fresh sessions of eight models choose the route for the next run after the scheduler's record of the same runs, after the memory that one of four memory-writing configurations wrote (a replay of the v1 memory pipeline of OpenAI's open-source Codex agent with its default writer models; ReasoningBank's released extraction instructions; mem0's open-source pipeline 2.0.19; Graphiti 0.30.2 with an embedded graph backend), after that memory beside the record, and after writer-free controls; each choice is scored against the route that the history's stated rules and costs require. The registered primary contrasts compare, per configuration, ended-case errors after the memory with those after the record. A second module tests prospectively, with a direction stated in advance, a contrast reported earlier on fresh histories of that earlier design; fixed earlier requests are re-sent as a drift check. The archive holds the registration (paper67/study5/REGISTRATION5.md), the design documents, the decision log, the study's code (which executes earlier frozen files of the author's Paper 7, unchanged in their bytes and checked at import against their freeze manifests), the two adapters with their fidelity records and pins, the development pilot's stamped gates, summary and report (its answers after the memories are sealed and not included), the spend projection, the calibration of the registered statistics, the registered histories, the anchors' frozen requests and original answers (from the author's public run records of Paper 7 Study 2), the pre-freeze reviews with their dispositions, and the chain scripts. MANIFEST.sha256 lists every file with its SHA-256; MANIFEST.sha256.ots is its OpenTimestamps proof. Manifest SHA-256: f4ed11b4c135a4f4eab16b2dc519566fc303b81c908299ef4bd4faa3a04bca42. Zip SHA-256: 4e04323beac7f481b7a4f3c418be5ae85124d59531167c3f51c17c123840005d. No confirmatory result is included. Licences: paper67/study5/LICENSE.md.