AI signal evidence questions
Answers about automation, holdouts, backtests and cryptographic records.
Is every trading algorithm AI?
No. A fixed rule can be systematic without being trained ML. Identify the actual decision component and its documentation. A bot routing somebody else's message does not prove the message came from a learned model.
Does a clean date split validate a model?
No. The worksheet checks one calendar condition and the user's declared test status. It does not inspect data, detect all leakage, measure predictive skill or estimate trading results. Every export remains untested.
Can a backtest be out of sample?
Its evidential status depends on the frozen experiment and what influenced development. If results shaped later choices, that period is development evidence. Historical simulation also remains distinct from messages delivered or positions executed in real time.
Does a receipt prove AI accuracy?
A verified commitment concerns its covered bytes and attestation. It does not identify model weights, training data or omitted predictions unless those are separately documented. Performance and completeness require their own evidence.
Sources
- QuantConnect optimization parameters
- ALFRED real-time and vintage data
- OpenTimestamps protocol documentation
Research guides
AI data leakage and holdout audit
Audit feature availability, outcome overlap and test exposure with a frozen experiment record and a worked calendar-boundary example.
Backtest, forward and live evidence
Reconcile simulated decisions, prospective instructions and account fills without promoting one evidence layer into another.
Algorithmic versus AI signal claims
Map fixed rules, trained predictions, AI explanations and order routing to the specific consumer evidence each needs.
Does a high accuracy percentage establish useful AI?
Not alone. It requires a defined event, horizon, complete population and appropriate baseline. A mostly one-class sample can give a high correct fraction to a constant rule. A ranking measure also differs from probability reliability. The class-balance and calibration chapters work through these questions using labelled fiction, without assigning toy scores to actual vendors.
Does an unseen checkbox verify a holdout?
No. It records the reader's declaration. Once outcomes influence development, another filename or model version does not erase that exposure. The checker examines only its simple calendar condition, not a dataset or access history. Its valid export remains untested even when dates are separated and the declaration says unseen.
Are research software, signals and copied accounts interchangeable?
No. A training framework, explanatory service, original forecast and follower account require different records. Product documentation can establish a capability without certifying a chosen strategy. The provider comparison therefore keeps task fit and missing evidence visible, with no fabricated universal financial winner or customer-review count.