01 / Define
Write the rules before seeing the result.
Start with one question. For example: does a particular pattern appear consistently in the historical observations you are allowed to use? Specify the sport, market, observation period and exact condition you want to investigate.
Record when the condition would count as present, when it would be ignored and which information would have been available at that moment. A rule that depends on knowing the eventual outcome cannot be tested as though that outcome was known in advance.
Keep the first version. If you change a rule after looking at results, save a new version and explain why.
02 / Check
Know what the observations represent.
Record the source, timestamps, collection method and whether observations are delayed, historical, synthetic or incomplete. Confirm that you have permission to use and store the data for your purpose.
In sports markets, the scheduled event time and the time an observation was available are different things. An incomplete sequence can hide what happened between two observations. Label those limits clearly in your notebook.
Never describe synthetic examples as real market evidence, or a retained old snapshot as a fresh market update.
03 / Explore
Read more than the headline return.
A backtest applies defined rules to a past dataset. It helps you inspect behaviour under those assumptions. It does not establish that the same behaviour will recur.
Review the number of observations, losing examples, largest decline, missing data and concentration of results in a few events. Write down how costs, timing and simulated execution were represented. Results that assume every desired action succeeds need that limitation stated beside them.
Ask whether a small number of examples dominate the result. A sports-season change or a different set of events may reveal behaviour that the original sample never contained.
04 / Challenge
Keep some data out of the design process.
A holdout is a separate portion of data reserved for evaluating rules after their initial development. For time-ordered research, preserve chronological order so later observations do not leak into earlier decisions.
Freeze the rules before reviewing the holdout. Once its results influence your next changes, that data has become part of development. Repeatedly tweaking an idea against the same holdout weakens its independence.
Walk-forward research repeats development and evaluation across chronological windows. Record which window supplied each result and which version of the rules was used. Multiple windows help expose variation; they do not guarantee future performance.
05 / Practise
Observe what happens next in paper mode.
Forward-paper research records how fixed rules behave on subsequent observations without submitting real orders. Save the condition, observation time, simulated decision and eventual outcome when it becomes known.
Keep skipped cases and missing observations as well as successful examples. A journal is most useful when another review can reconstruct what was known at the time.
Rule version · sport and market · observation timestamp · data source and delay · reason for the simulated decision · execution assumptions · outcome · limitations.
Paper execution cannot reproduce every constraint of a real market. Treat the difference as an unresolved question, not a reason to promote a system automatically.
06 / Review
Decide what you have learned.
Compare the original question with the evidence. Identify which assumptions held, which failed and which remain untested. Record unsuccessful ideas so they are not quietly removed from the story.
A useful next step may be more data, a clearer rule or abandoning the hypothesis. None of these research stages requires real-money trading.
The Edge Ladder private preview provides guided strategy drafting and research workflows. Explore the workspace overview for current capabilities and read the questions and answers for features still being prepared.