Skip to system research
Build A Trading SystemRules / Evidence / Replay

Questions before relying on a review

Educational research design and invented arithmetic examples. No strategy returns, broker execution or production certification are supplied.

Original edition: 8 October 2026. Source limits appear below. Expanded edition: 13 October 2026.

Educational research design and invented arithmetic examples. No strategy returns, broker execution or production certification are supplied.

Where should a trading-system project start?

Write an observable hypothesis and data-availability contract before selecting indicators. Specify the instrument, clock, entry, exit, sizing, costs and rejection rules. The rule-sheet workbench captures those decisions without running a backtest or making a performance claim.

Does a passing software test validate a trading edge?

No. Fixtures can check timestamps, formulas, accounting and order-state logic. They are not independent market samples. Strategy evidence requires its own declared evaluation design, execution assumptions, costs and limitations.

Can I revise the model after seeing the holdout result?

You can research a revision, but the inspected holdout is no longer untouched evidence for that revision. Preserve the original result and change log. Do not repeatedly tune on the same reserved period and describe every later pass as a fresh independent test.

Can production rely on my laptop staying online?

Not under this workspace's production policy. Scheduled refreshes, publishers, monitors and operational artifacts belong in managed cloud services. Local research and temporary QA are allowed, but a local process is not a production fallback or deployment certificate.

Sources and scope

Why publish a negative synthetic result?

Because the task is to make the mechanics reproducible, not select a flattering demonstration. The fixed data and policy produce positive and negative simulated positions whose total is negative 1.20 after the declared fees. That number is a software-reference outcome, not a historical loss observed in a market. A later change to data, timing, fees or exit creates a new fixture identity and should retain the previous one. Erasing the old result would undermine the teaching record without supplying an economic hypothesis or independent validation.

Does a file match prove an untouched holdout?

No. SHA-256 equality identifies bytes of the selected artifact. It cannot establish who viewed a period, when a candidate was selected or whether a later change used its outcomes. Access and decision logs are different evidence. The new inspector explicitly does not rerun the engine or authenticate a broker result. A complete manifest can be useful while its selection history is missing, and a copied receipt is not proof of a new execution. The reference packet's classification stays synthetic even when every uploaded file exactly matches it.

Is this the proprietary instrument research factory?

No. This is an educational site with a library mechanics fixture. It does not replace canonical routing, training objectives, resource bounds, native exits or the separate independent master. It also does not change the established observation-window fill convention. The tutorial's next-open fills are declared for its own invented data only. A real instrument task needs the full authorised process; software components and positive arithmetic cannot silently substitute for it. Cloud publication of the website is not model promotion or certification of an account route.