spb/wp13_uqo

1% ============================================================================2\section{Conclusion}3\label{sec:conclusion}4% ============================================================================56This paper has described VQuant, a deployed platform that turns a natural-language request into7a quantitative financial report under one organising principle: the language model plans and8synthesises, and deterministic code retrieves data and performs computation. We specified four9mechanisms as implemented and audited against the source code---a recursive, batched, streaming10tool-calling loop over a 245-tool catalogue; a three-stage context-management procedure with its11actual constants; a single-call computation convention and the process-isolated sandbox that12executes it; and a document pipeline from one Markdown artifact to PDF, DOCX, Beamer, and13live-formula spreadsheets---and we corrected the earlier description of the system where the14code differs from it. In the vocabulary of design science, the paper contributes a situated15implementation and a demonstration; the evaluation step is specified but not performed.1617The limits of what has been shown are therefore clear. The architecture removes numbers from18the model's remit, but whether this reduces unsupported numbers to near zero, by how much the19single-call convention reduces round trips and at what cost in failed executions, how much the20summarisation procedure loses, and what the design costs relative to simpler agents are empirical21questions that the protocol of Section~\ref{subsec:protocol} is written to answer and that this22paper does not. Executing that protocol is the natural next step and the one on which any claim23of superiority must wait. Three engineering changes are also warranted by the audit: a bound on24recursion depth, a per-model price table in the ledger, and the restoration of the pre-built25numerical routines to the tool catalogue. Beyond these, the design admits extensions that the26protocol could also evaluate: a verification pass in which a second model audits the first's27narrative against the computed numbers, provider-independent tool exposure through the Model28Context Protocol, and generalisation of the deterministic-computation substrate to other domains29in which auditable numbers matter. We offer the specification, the audit, and the protocol as the30contribution, and the pattern they describe---a language model as a disciplined orchestrator of31trustworthy tools rather than as an unverified oracle---as a hypothesis to be tested.32