Deterministic computation
Backtest, risk, and statistical metrics come from reproducible engines. Agents organize and explain.
Trust
Trust is an operating boundary
For quant teams, agents need to be reviewable, constrained, and replayable. DQT emphasizes operating controls, not exaggerated performance or autonomous trading claims.
Permission boundaries
Agents call only approved data, tools, and workspaces, avoiding scope creep.
Human approval
Critical steps require owner review, especially around live, risk, or public-facing actions.
Audit records
Prompts, inputs, tool calls, parameters, results, and approval state remain attached to each run.
Paper/live separation
Research, paper validation, and live workflow support remain clearly separated.
Careful claims
No performance guarantees, fabricated customer proof, or data-vendor positioning.
Public boundary
Describe the process, not sensitive details
DQT can describe controls, operating state, permissions, and auditability, but must not reveal strategies, signals, parameters, internal schemas, or contract details.
trust_boundary:
show: process_controls
show: agent_boundaries
hide: strategy_logic
hide: internal_data
hide: client_terms