RESEARCH LOG
The trail matters.
MSRLab is being built in public from the point where the record is still small. These notes preserve how questions, systems, and standards evolve.
Questions change.
The evidence stays traceable.
From trading gurus and perfect EAs to a research portfolio
The mindset shift is from marketed certainty and one-shot systems to testable claims, complementary return drivers, live feedback, and continuous maintenance.
NOTE ↗02
Why strategies are rejected: an MSRLab taxonomy
A strategy can fail because its premise is weak, its evidence is fragile, its economics are unrealistic, or its portfolio role is redundant. Rejection records which category failed and why.
NOTE ↗03
Why MSRLab is human-directed—not fully autonomous
Agents can widen the search, implement faster, and challenge more aggressively. Judgment about evidence, risk, and capital remains human-governed.
NOTE ↗04
Portfolio thinking versus the search for one perfect EA
A moderate edge can be valuable when its return driver complements the rest of the book. A spectacular backtest can be useless when it duplicates existing risk.
NOTE ↗05
Why live deployment begins another research phase
Backtests define an expected profile. Forward and live behavior reveal execution, regime, and attribution differences that become the next experiment.
NOTE ↗06
What the PBZ variants did not preserve
Repeated filtering reduced bad baskets but often removed the behavior that made the original approach interesting. That failure narrows the next question.
NOTE ↗