Process
Every system moves through the same sequence in the same order. The stages exist to make a strategy fail early and cheaply, and the fourth one exists because a profitable backtest that quietly used future information is worse than no backtest.
What the database holds
These are counts from the platform database. The research programme behind these systems tested substantially more configurations than are recorded here — those figures are described on the research page, and they are deliberately not restated as computed metrics until their source logs are ingested and can be recomputed.
Systems tracked
Paper3
Backtest runs ingested
Will be BacktestedNo backtest trade log has been ingested. The research artifacts that produced the published figures are not present on this machine.
Backtest runs ingested: not yet available. No backtest trade log has been ingested. The research artifacts that produced the published figures are not present on this machine. Will be graded BACKTESTED.Closed trades ingested
Will be PaperNo performance export has been ingested yet.
Closed trades ingested: not yet available. No performance export has been ingested yet. Will be graded PAPER.Automation events
Will be PaperNo terminal logs have been ingested.
Automation events: not yet available. No terminal logs have been ingested. Will be graded PAPER.Stage by stage
A hypothesis about a market mechanism, written down before any data is touched. Stating it first is what makes it falsifiable later.
The idea is tested across a wide instrument universe rather than the one instrument that inspired it. Most ideas die here, and an idea that survives on only one instrument is published as such.
A full historical simulation with real spreads. This produces the first numbers, and they are the least trustworthy numbers in the process.
A mandatory gate. Every indicator, every signal and every fill is checked for information that would not have been available at the moment of the decision. A strategy does not pass this stage by being profitable; it passes by being checkable.
The headline result is corrected for the act of having selected it. A configuration chosen as the best of many is biased upward by the choosing, and the corrected band — not the headline — becomes the planning number.
Execution on a demo or evaluation server against live prices. Real spreads, real slippage, real execution failures, simulated capital. Results from this stage are graded PAPER.
Real capital, with an equity kill-switch and position sizing aligned to the account's rules. Results from this stage are graded LIVE.
Continuous comparison of live behaviour against the validated expectation. Divergence is a reason to reduce or halt, not a reason to re-fit.
A strategy is retired when its edge is gone or its regime has changed. Retirement is a normal outcome and is recorded rather than quietly stopped.
Where each system sits
Stated stage · live
Will be PaperStated stage · paper forward
Will be PaperStated stage · validated
Will be Paper