Skip to model research
Systematic Trading ModelsSpecification / Evaluation / Release

Trading tools and source notes

Libraries, platforms and proprietary offers need different assessments. Entries describe documentation and missing evidence, not invented model scores.

Original editorial revision: 9 October 2026. Expanded edition: 14 October 2026.

Research design and arithmetic, not model recommendations. Classification scores are not monetary returns; a registered artifact and a passing software test are not independent strategy validation.

Different jobs, different evidence

Request immutable model identity, training and label periods, feature availability, transformation fit scope, action and exit policies, complete trial registry and independent evaluation. A purchased signal record additionally needs its original instructions, amendments and execution assumptions. Ask what the documented tool does before comparing it with a proprietary model.

Systematic Model Release Workbench

Review a model as a versioned decision system: data identity, entry objective, exit ownership, reserved evaluation and durable operational state.

Different resource roles, not a unified tournament

An alerting platform, scanner, news feed, copying service and signal publisher contribute different objects to a model workflow. They cannot be ranked as if they submitted the same identified candidate to the same independent experiment. The directory records those roles and points to official descriptions. An alert can carry a decision, a scanner can define eligible candidates and a news feed can supply features; none automatically discloses a complete model. A copying platform adds another receiving-account policy. The resource profiles preserve those distinctions rather than manufacturing comparable financial scores.

The minimum comparable model contract

A meaningful comparison needs the same label definition, evaluation population, causal source scope and accounting convention, or an explicit explanation of why they differ. Identify each candidate's trained parameters or deterministic rule and selection history. Keep preprocessing, threshold, exit and data revisions in the version record. Then explain which outcome is measured. Binary precision and net account return cannot share one unexplained score column. If the required records are absent, a directory can still be useful for finding mechanisms, but its missing evidence is not evidence that one vendor won a trading contest.

Ask about the receiving side

A source model can produce the same instruction for two users while their account path differs. Orders, quantity settings, broker availability, currency and synchronization belong to the receiving policy. A publication timestamp does not measure that path. No profile in this directory is backed by our own paid subscription, captured order stream or customer-cohort study. The new workbook's examples are invented and platform-neutral. They teach specification questions without suggesting that any vendor generated the packet. A provider's public documentation establishes its declared functionality, not our authentic execution of it.

Model-card and probability references