Classification metrics versus trading value
Read accuracy, precision, recall and class imbalance without confusing a correct label with a profitable executable decision.
Write what would reject the candidate before selecting parameters. A flexible story that can explain every result is not a bounded hypothesis. Separate software tests, causal-data checks and independent strategy evaluation.
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.
Write what would reject the candidate before selecting parameters. A flexible story that can explain every result is not a bounded hypothesis. Separate software tests, causal-data checks and independent strategy evaluation.
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.
Declare the source, causal information clock, transformation, target, fitting rule, action, exit, sizing and costs. Choose the independent evaluation boundary before looking at its outcome. Preserve unsuccessful experiments and later amendments.
Documentation establishes the stated mechanism, not tested trading performance. Checked 9 October 2026; a later version can differ.
Read accuracy, precision, recall and class imbalance without confusing a correct label with a profitable executable decision.
Audit availability times, train-only transformations and forward targets before trusting a trading-model split.
Keep candidate selection, independent strategy validation, software checks and cloud production state as separate durable records.
A testable hypothesis defines what would make it fail and which untouched evidence may answer the question. An entry rule that can be redefined after every unfavorable label is not bounded. Neither is an exit tournament opened after inspecting the reserved result. The fictional revised-contract card shows a new version needing new evidence rather than inheriting the old favorable score. Failure conditions can concern causality, population, economic result or operations, but must retain their own scope. A data-arrival defect is not the same as a negative account return.
Software tests deliberately contain known results: a default Brier loss, an equality threshold or a prohibited boundary-crossing label. They should catch incorrect code. Independent strategy evidence cannot simply reuse those answered fixtures and call them blind research. Preserve the canonical factory's full completion and master requirements for a real instrument task. A verified no-model master can be a valid end state; an unimplemented master or a compact intermediate grid is not that end state. This site describes the distinction without initiating actual model research.
A source-causality defect, incorrect metric formula and absent deployed route reject different assertions. State which one failed and which evidence remains valid. An operational missing file cannot by itself establish that an entry hypothesis failed its full independent financial master. This scope discipline makes adverse findings useful without overstating them.