Maintaining investment records

An investment office repeatedly encounters statements, notices and instructions that change the same portfolio. This report tests whether persistent records and shared processing reduce repeated work while preserving separate investor states.

A paired experiment processes the same synthetic investment histories separately and together, with and without persistent records and validation. Records reduce repeated retrieval. Shared processing has lower recorded inference costs, while the tested improvement saves more in separate processing. Successful primary contexts are at the scored interpretation ceiling, and most exact-output improvement is a change in how abstention is represented. The report preserves failures, costs, prompts and receipts. It measures a component of processing cost, not investment returns or the total value of an investment office.

What was measured

The experiment uses synthetic investment histories and compares separate and shared processing. It records inference costs, interpretation scores, exact-output behavior and service failures.

Persistent records reduce repeated retrieval. Shared processing has lower recorded inference costs, although the tested improvement saves more processing cost in separate service. Among successful primary contexts, interpretation scores reach the measured ceiling; most exact-output improvement concerns how abstention is recorded.

The PDF gives the experimental design, scoring rules, outcomes and limitations. Its results are unchanged from the accepted 5 October edition.

Interpretation

These are measurements of a document-processing task. They do not establish the full cost of an investment office, reliable autonomous execution or investment returns.

The next empirical questions concern performance through an entire workflow: keeping commitments current, detecting conflicting instructions, investigating relevant possibilities and recognizing when a decision needs approval.

Publication details

Separate Records, Shared Processing. John Komkov. 5 October 2026. This download preserves the manuscript supplied for the 8 October 2026 website snapshot.

John Komkov has commercial interests in Argosy and Megalith. The report describes its AI methods, provenance and scoring limits.

The Formation Game studies the arrangements that make shared investment possible.