Make the experiment inspectable.
Record datasets, assumptions and evaluation conditions. Context belongs beside a result, not in a forgotten footnote.
02 / THE RESEARCH METHOD
Every idea needs
a paper trail.
A disciplined process keeps the questions, assumptions and evidence close together. Explore the principles behind our research workflow.
QUESTION 01 / OBSERVE
What did we actually know?
Point-in-time context comes first. Features must reflect information available at the decision time, not knowledge introduced later.
RESEARCH PRINCIPLE
Feature timestamp ≤ decision timestampFIELD NOTE
Record the source, the time and the transformations. Future labels belong to offline evaluation, not the inputs available to a strategy.
QUESTION 02 / TEST
What are we assuming?
Make timing, costs and simulated fills explicit. Use the same experiment boundaries when comparing a candidate with its benchmark.
RESEARCH PRINCIPLE
Candidate and benchmark share evaluation conditionsFIELD NOTE
A plausible story is not a result. Reproducibility means another run can be traced back to the same data and assumptions.
QUESTION 03 / CHALLENGE
What would disprove the idea?
Look beyond the fitting period. Critique the evidence, inspect gaps and make uncertainty visible before treating a hypothesis as worth further work.
RESEARCH PRINCIPLE
Out-of-sample evidence is distinct from training fitFIELD NOTE
Negative findings are useful. A weak or contradictory result should change the research decision, not disappear from the record.
QUESTION 04 / REVIEW
Who has the authority?
Human review and explicit eligibility boundaries sit between research and a paper experiment. A recommendation does not itself authorise execution.
RESEARCH PRINCIPLE
Research recommendation ≠ execution permissionFIELD NOTE
Paper activation remains bounded by portfolio state and risk gates. No part of this illustrative notebook sends an order.
CONCEPTUAL WALKTHROUGH / NO MARKET FEED / NO PERFORMANCE FIGURES
RESEARCH CULTURE
Keep the questions.
Not just the wins.
Treat criticism as part of the work.
Source gaps, contradictory claims and failed hypotheses help refine the next question. They are not obstacles to hide.
Separate evidence from authority.
An interesting research finding still has to pass review. An intelligence agent does not gain execution permission by sounding confident.
THE NEXT CHAPTER