Research / A closer look
Turn a question into a test. Use the result to decide what to keep and what to discard.
01 / The process
What would have to happen for me to be wrong? I want a test that can challenge the explanation, including the way I measured it.
See the decisions and outcomesWhat could create the opportunity? Who takes the other side, and why might it persist?
Write the claim, comparison, and failure conditions. Separate exploration from a test specified in advance.
Use information available at the time, realistic costs, and reproducible code. Check other periods and challenge selection bias.
Keep, revise, discard, or leave unresolved. A promising backtest still needs to hold up in execution.
02 / Questions about the market
Capital is finite and nobody sees the whole picture. These guide an investigation; a trading rule still needs evidence.
01 / Market depth
Available size is finite. A sale can fill at more than one price.
Illustrative bids: 2 units at 100, 3 at 99.
Each dark block is one unit.
Available buys: 2 at 100, 3 at 99.
A fixed book, without new or cancelled orders, hidden liquidity, latency, or fees.
About execution costs ↗02 / Different horizons
Participants act over different horizons, with different information and different constraints.
03 / Interaction
Orders and attention can converge. A shared level is a question to investigate before anyone calls it an edge.
04 / Reflexivity
Traders react to what the market shows, and their reactions change what it shows next. A pattern can fade because people trade it.
03 / AI, code, and judgment
AI helps me explore and build. Code measures. I decide what the evidence supports and bear the consequences.
Explore
Use AI to locate sources and turn an idea into an implementation.
Challenge
Review assumptions, data handling, and costs. Agreement between models does not settle the question.
Decide
Say what the evidence supports and what it does not. Preserve the reasoning and the limitations for the next revision.
An appealing historical result led to harder questions: transaction costs, correlated losses, assets that disappeared, and whether simulated fills could happen in practice. Review objections led to amended tests and unresolved limitations.
That is the purpose of iteration: improve the question and the measurement, while keeping a research result distinct from live validation.
Read my thoughts on AI and judgment ↗04 / Tools in use
What it measures and where it stops being useful.
Entry conditions, invalidation levels, and the events that could change the plan.
My internal system for research and analysis, developed with Claude Code.
Code and the history of its revisions.
Behind Emergence
I research and trade. I share the work so you can challenge it and build your own judgment.
Why Emergence exists