How should AI call a backtest tool—and what should a researcher check?
Twelve synthetic bars and two specified signals expose inputs, execution and costs in a real engine invocation.
AI structures the request, the engine calculates, and a researcher checks. Both synthetic trades lose money. The run demonstrates mechanics, not strategy validity or profitability.
Define the calculation
Check execution and costs only. Signals on bars 2 and 7 go long and short respectively, enter at the next open and exit two bars later. Each side costs 5 bps. All 12 bars are predefined synthetic teaching inputs, not a selected market window.
Inspect the actual output
BarBacktester returns a long trade from 102 to 100 and a short trade from 98 to 99. Round-trip costs are 0.1% each. Net returns are approximately -2.0608% and -1.1204%. The public JSON retains inputs, trade records, engine-file hash and run time.
Review the mechanics
Match each signal to the next open, check exit timing, short direction and fees. This can check whether a calculation followed the request; it cannot establish data authenticity, robustness or future profitability.
Separate signals, sizing and risk
A signal defines a condition, sizing defines exposure, and risk rules reject or pause activity. AI can help specify each part, but sources and code still need review. Narrative must not fill gaps in evidence.
State what was not tested
Two synthetic trades support no parameter-search, market-performance or out-of-sample conclusion. No account is accessed and no orders execute. Public MCP is in development; this is a real local-core invocation, not an available remote service.
Questions
Is this an actual invocation?
Yes, using explicitly disclosed synthetic teaching inputs. The screenshot presents that run; its JSON can be inspected.
Can two records validate a trading strategy?
No. This is not evidence of validity or future returns.
Sources and scope
External sources support the definitions and method discussion; they do not validate WeiMind’s research results.
Courses teach methods and tools; signal subscriptions have distinct scope and eligibility. Backtests and past results do not assure future returns. No execution or automatic copy-trading. Interest registration only.