Sig memory profile

Type: agentic-system-analyses/types/agentic-system-memory-profile.md

Comparison rationale

The accepted records describe a memory design, not inspected application wiring. All positive findings therefore retain a claimed basis. Reading every release document does not resolve the unavailable application loop. No axis asserts a complete inventory merely because one local part is classified.

The two duplicate private-object supersessions in the supplied reconciliation resolve the epistemic factual and personal objects to MEM-OBJ-factual-record and MEM-OBJ-personal-layer. Their epistemic route findings still apply. The organizational proposal EPI-OBJ-org-version remains distinct from MEM-OBJ-team-knowledge: human selection and approval intervene. No unchanged-content identity is assumed.

The lineage findings distinguish authoring the input, Sig compiling a derived private account, preparing a proposal, and importing approved material or skills at another installation. Other-compiled names a derivation path, not a claim of faithful compression. The indeterminate semantic relation recorded by EPI-RTE-private-synthesis and EPI-RTE-org-draft prevents choosing a curation operation. The product's word synthesis is not evidence for the controlled value synthesize.

Direct human reading supports a local pull route. It does not settle whether later AI context is retrieved on request or automatically injected. EPI-RTE-context-use supports claimed knowledge force at intended later consumers, while retaining unknown selected content and personal-layer provider inclusion. Synchronization and immediate availability establish no push selector. Likewise, the explicit human decision in EPI-RTE-share-approval classifies publication admission, not local private admission, recipient review, or a retained validation/enforcement object.

Trace learning remains unresolved independently of advertised AI benefit. Neither user-initiated capture nor Sig's physical file write settles local admission control. No original recorded trace source and qualifying automatic update chain are established. The records do not provide bounded absences for skills, optional integrations, transcripts or provider context. Those branches cannot be dropped to produce a negative or an inapplicable trace-source axis.

Diagnostic linkage, telemetry and public support intake are excluded from this memory boundary rather than classified as app context. Their records identify no admitted repair or later app-change consumer: MEM-OBJ-install-id, RT-RTE-analytics-egress and RT-RTE-product-feedback. This exclusion does not claim they lack storage or diagnostic use.

Source-read log: none. Classification used only the supplied accepted members, reconciliation and governing contracts; no source ambiguity required a new read.